How we classify changes
- Corrected: the page previously stated something that was wrong or has become wrong. The substantive text is changed and the change is logged.
- Updated: new material facts have been added. The prior text was not incorrect at time of writing.
- Improved: structural, technical, or editorial change with no effect on factual claims.
- Retracted: a claim has been withdrawn. The original wording is preserved and struck through rather than deleted.
- Withdrawn: a planned project or commitment was cancelled before it produced any published result.
We do not silently edit factual claims. A correction to one page is logged as its own entry. A bulk review pass, in which many pages are checked and edited at once, is logged as one aggregate entry that names the pages or groups of pages and the kinds of change; the individual edits in such a pass are not listed one by one. Entries in this log may be edited after publication to correct or narrow their wording, and such edits are noted in the log. We aim to publish a confirmed correction within 48 hours. Company profiles carry a last-checked date and show the date their next review is due. See the editorial policy for our sourcing standards and editorial commitments.
Change log
| Date | Page | Type | What changed |
|---|---|---|---|
Policies and market timeline | Clarified | The editorial policy and legal notice said we aim to reply to correction requests within five working days, while the corrections page said 48 hours. All now say we aim to acknowledge a request within one business day and to publish a confirmed correction within 48 hours. The cookie policy now lists every measurement the site sends, including printing, script errors, missing pages, opened sections and the pack chosen when registering interest. The privacy policy now states the salted hashes kept for up to 2 and 62 days to limit form abuse. The market timeline left out profile events recorded with a numeric date, which hid 67 events from September and October 2026; they now appear. Descope, Grip Security and Britive entries now carry exact dates and their own sources. | |
Profiles, comparisons and reference pages | Corrected and clarified | A red-team review corrected: Cremit's March 2026 rotation feature covers the tokens Cremit uses for its own scan integrations, not customer API keys; Infisical documents machine identities that work like service accounts, so its service-account row now reads Documented by vendor; SPIRE implements the open SPIFFE specification (it is not itself a standard); Traefik's 3.5 billion figure counts Docker image pulls; the consolidation tracker counts 30 deals; profile timelines are now in date order. The directory's relevance column is renamed NHI scope and defined in the methodology. A page for companies explains how records are made and corrected. | |
Company profiles and 39 reference pages | Re-checked and corrected | All company profiles were re-checked against company, investor and filing sources. Corrections: Keyfactor closed its acquisition of Cofide on 10 September 2026 (our records said completion was not confirmed); Nudge Security's announced funding is about $39M, not about $30M; MintMCP renamed itself Mintive on 12 October 2026; the frameworks hub described one Cloud Security Alliance survey where there are two plus a guidance paper. The publisher disclosure was re-checked against HumanAudit's company plan: HumanAudit has retired its NHI exposure scan and builds and sells no NHI scanner and no NHI discovery, monitoring or security tool. An interim version of this disclosure mentioned a free NHI exposure scan on humanaudit.ai; that offer has been withdrawn. Timelines gained dated events from September and October 2026. Pages whose facts could not all be re-read from primary sources keep their earlier check date. | |
Site-wide (about 270 pages) | Corrected and updated | A page-by-page review checked titles, opening answers, tables, wording and sources, and re-checked dated facts against primary sources. Factual corrections: C1 (formerly ConductorOne) is now recorded as subject to an acquisition agreement with Workday, announced 13 October 2026, and a news item was added; ISO/IEC 27090 is recorded as published (14 October 2026); the Cloud Security Alliance has published two NHI survey reports, not one; the WIMSE practices draft was submitted to the IESG on 30 June 2026, not in September; SPIFFE graduated in the CNCF on 23 August 2022 (announced 20 September 2022); the CSRB count of 22 organisations in Storm-0558 is attributed to the CSRB report, not to Microsoft's July 2023 disclosure; SailPoint describes Entro as available to SailPoint customers as standalone offerings; Rubrik's Identity Continuity is in private preview and Strata had a product of the same name; Idira is described as a Palo Alto Networks platform with upgrade paths for CyberArk SaaS customers, not as a relaunch of CyberArk; the CircleCI incident report is dated 12 January 2023; statements that the OWASP NHI Top 10 is vendor-neutral were removed; the classification axes are credited to Abhishek Singh and Avadhoot Patwardhan of Oleria. Breach pages gained incident records, ratio figures now state sponsor, method and sample, notes about material combined from nhimanagement.com moved to each page's history, and claims.csv was repaired (its editor column had split rows). | |
/, /about-humanaudit.html, /privacy-policy.html, /contact.html, /methodology.html, /companies/directory.html, /nhi-governance-pack.html | Updated | A monthly change log sign-up was added, with double opt-in; the privacy policy describes it. A page about HumanAudit Inc., the publisher, was added. The empty classification-axes frame, with its disclosure that the axes were developed with input credited to a person at a profiled company, moved from the company directory to the methodology. Pages consolidated from nhimanagement.com now carry the label "Sourced guide" and keep their migration note in a page history at the end. Company profiles show the long research sections collapsed; no facts were removed. | |
/privacy-policy.html, /cookie-policy.html, /contact.html, /corrections.html, /nhi-governance-pack.html, /robots.txt | Updated | Analytics now depend on the time zone your browser reports: in the EU, the EEA, the UK, Switzerland, Quebec, Ontario, Brazil, India and Korea, or when the time zone is UTC or unknown, nothing is loaded until you accept; elsewhere analytics run from the first page with a notice and a one-click opt-out. Privacy signals are honoured everywhere. Event-level analytics data is now kept for 14 months. Three forms were added (register interest, contact, correction request). They send email through Postmark and keep a copy outside the public website for up to 24 months. The privacy and cookie policies describe both changes and took effect on 10 October 2026. | |
company profiles, /data/, /governance/, /learn/non-human-identity-lifecycle.html, /research/nhi-ratio-benchmarks.html, /use-cases/, /glossary/, /frameworks/workload-identity-federation.html, /breaches/salesloft-drift-2025.html, /breaches/snowflake-unc5537-2024.html, /privacy-policy.html, /nhi-governance-pack.html | Corrected | Aggregate entry for a second pass on 10 October 2026. The editorial responsibility label on profiles and in the datasets now reads "Akshay Dubey (editor), HumanAudit Inc." instead of "Akshay Dubey / HumanAudit Team", because one editor is responsible and "Team" suggested more reviewers than exist. Pages migrated from nhimanagement.com were checked again: the ratio benchmark page credited the 82:1 figure to Palo Alto Networks, but it comes from CyberArk's April 2025 edition, a different sample; the credential rotation page said federated credentials live for minutes, while the AWS and Google defaults are one hour; the workload identity federation page called federation the mechanism behind zero standing privilege; the breach case library had wrong CircleCI revocation dates; several glossary entries had unsourced or imprecise wording. These were corrected, and the "under review" label was removed from the pages that were checked. The breach case library keeps the label until four timeline facts have primary links. The Drift case said refresh tokens were extracted; Google's threat intelligence report says OAuth tokens were compromised, and the page now says that. In the Snowflake timeline, the Live Nation and Santander rows now say they are not linked to UNC5537 in any primary source we found. Six pages that repeated other pages were merged, each with a permanent redirect: ownership and attribution into the new owner assignment guide, IAM vs NHI into the IGA, PAM and CIEM guide, the provisioning and decommissioning playbooks into the lifecycle guide, service account hardening into the service account governance guide, and the management checklist into the management guide. The governance hub became the NHI governance pillar, with an 8-stage lifecycle model (v2), and new pages were added on audit readiness, an evidence catalogue, PCI DSS v4.0.1 Requirement 8.6, 47-day TLS certificates, owner assignment, least privilege, non-person entities and vendor transparency. Five core pages are now available as machine-assisted translations in German, French and Japanese; the English pages are authoritative. The privacy policy (now effective 10 October 2026) has a new section on translations and on the optional Google Translate links in the language menu. | |
company profiles, /companies/compare/, /data/, /about.html, /contact.html, /editorial-policy.html, /nhi-governance-pack.html, /privacy-policy.html, /terms-of-use.html, /legal-notice.html, /breaches/okta-support-2023.html, /learn/what-is-non-human-identity.html, /learn/non-human-identity-security.html, /learn/non-human-identity-management.html, /learn/service-account-governance.html, /learn/certificate-lifecycle-management.html, /learn/nhi-discovery-methodology.html, /frameworks/dora-nhi.html, /index.html, /site.webmanifest, /robots.txt, /llms.txt | Corrected | Aggregate entry for an independent audit pass on 10 October 2026. Withdrawn: the statement that HumanAudit's products do not compete with the tools profiled here, because some planned documents overlap with services some profiled companies sell (the Encryption Consulting profile now says so); and the description of the planned RFP kit as vendor-neutral, replaced by what the kit will not do. The web app manifest still called the site vendor-neutral and was corrected. Company profiles that said a capability "is not documented" now say that public documentation reviewed for the profile did not establish it, and the profile research note says "sources checked" instead of "source audit". The Okta case study stated as fact that no technical control distinguished a machine credential from a personal password; it is now labelled as our reading of Okta's public account. The Veza 17:1 ratio now states Veza's described method. The security primer used a different OWASP root-cause grouping from the OWASP page and treated keys and certificates as identities; both now match the site's taxonomy. The management guide and the metrics page gave different counts of metrics; they now agree. The service account guide wrongly said Kubernetes service accounts are covered separately in the taxonomy. The certificate guide said a team could be an owner; it now asks for a named person. Added: Articles 20 and 21 of Delegated Regulation (EU) 2024/1774 to the DORA page. The privacy policy now names the transfer safeguard for Google Analytics and lists the rights to restriction and portability; the terms and the legal notice now match the CC BY licence on the datasets. | |
/companies/ (146 profiles and the comparison pages), /index.html, /disclaimer.html, /legal-notice.html, /editorial-policy.html, /learn/index.html, /frameworks/index.html, /llms.txt, guide and glossary pages | Corrected | Pages described the site or its content as independent, vendor-neutral or neutral. Company profiles and comparison pages were headed "Independent editorial profile", profiles carried an "Independent research summary" note, guide and glossary pages said "This reference is free and vendor-neutral", and the disclaimer, legal notice, home page, two hub pages and llms.txt used similar labels. The publisher, HumanAudit Inc., was then described as developing a paid NHI discovery and ownership product, so these labels were removed (that description was itself replaced later the same day; see the entry on HumanAudit's description) and the disclosure is stated instead. The profile field "Company review" is now "Company input": 145 profiles record no company input, and the Oleria profile records questionnaire answers labelled Vendor-supplied. | |
/contact.html, /privacy-policy.html, /cookie-policy.html, /terms-of-use.html, /disclaimer.html, /editorial-policy.html, /methodology.html, /.well-known/security.txt and other pages that gave a contact address | Updated | The contact address is now hello@nhigovernance.com, replacing the previous humanaudit.ai address, on every page that gave one, including the legal pages and security.txt. The effective dates of the privacy policy, cookie policy, terms of use, disclaimer and editorial policy changed from 7 October 2026 to 9 October 2026 because their text changed on that day. | |
/data/index.html, /data/claims.json, /data/claims.csv | Clarified | The data page now states what the verified date on each claim row means: the date its company profile was last reviewed, applied to every claim from that profile. It is not a separate check date for each claim. Before this change the page did not define the date. | |
/corrections.html | Updated | The policy at the top of this page now says that bulk review passes are logged as aggregate entries, not one entry per edit, and that entries may be edited after publication. One entry dated 9 October 2026 was edited after publication to remove customer names. | |
/frameworks/eu-ai-act-identity-obligations.html, /frameworks/eu-ai-act-nhi.html, /frameworks/nis2-nhi.html, /frameworks/iso-42001-nhi.html, /frameworks/standards-tracker.html, /frameworks/owasp-nhi-top-10.html, /ai-agents/multi-agent-authorization.html, /ai-agents/agent-revocation.html, /breaches/microsoft-storm-0558-2023.html, /breaches/index.html, /breaches/okta-support-2023.html, /companies/clutch-security.html, /companies/lasso-security.html, /learn/non-human-identity-lifecycle.html, /learn/nhi-maturity-model.html, /learn/machine-identity-management.html, /learn/nhi-vs-machine-identity.html, /use-cases/nhi-discovery-queries.html, /methodology.html, /index.html, /llms.txt | Corrected | Aggregate entry for a review pass on 9 October 2026. The EU AI Act identity page was retitled "EU AI Act articles that bear on identity evidence (our reading)", because the Act names no identity obligation, and its Article 12 answer now says the article applies to high-risk AI systems. The NIS2 translation is labelled as our reading. The Storm-0558 pages said Microsoft withdrew its crash dump explanation; Microsoft's 12 March 2024 update corrected the claim that a crash dump containing the key had been found, and the Cyber Safety Review Board notes Microsoft did not recant the theory. The Okta page now links Okta's 29 November 2023 post. The A2A row said version 1.0 was announced on 9 April 2026; GitHub shows v1.0.0 released on 12 March 2026 and v1.0.1 on 28 May 2026, so the version 1.0.1 removed by an earlier entry is now confirmed and listed. The Clutch Security and Lasso Security profiles described SOC 2 as a certification; it is an auditor's attestation, and the sources show only badges. The EU AI Act page now counts two new Article 5(1) prohibitions. Two unsourced general statements on the ISO/IEC 42001 page were removed, and the service account governance page now says the Gartner research it mentions is paywalled and is not linked. The lifecycle page is consistently four stages, with the five-stage split labelled as a variant, and the lifecycle and maturity models carry a version line. Added: an RFC 7009 and RFC 7662 paragraph to the revocation guide, the 1,000-event limit of the CloudTrail example, a "Limits of this research" section in the methodology, and two non-vendor figures at the start of the home page statistics. | |
/about.html, /editorial-policy.html, /methodology.html, /legal-notice.html, /disclaimer.html, /index.html, company profiles, guide pages, /llms.txt, /nhi-governance-pack.html | Updated | HumanAudit's description changed to match its current company plan. HumanAudit describes itself as an AI assurance evidence company: it sells governance documentation and plans documentation packs for NHI governance (with an RFP kit that does not name, rank or recommend any vendor), AI agent governance and certificate readiness. It does not sell NHI discovery, monitoring or security tools. Earlier text said HumanAudit was developing a paid NHI discovery and ownership product and offered a free exposure scan; that text was removed. The guide-page box, the header link and the footer now point to a page that describes the planned packs, with no prices; company pages still carry no offers. The disclaimer now states that HumanAudit is not affiliated with ISO, IEC, OWASP, the Cloud Security Alliance, the PCI Security Standards Council, the CA/Browser Forum or any certification body, and the corrections policy now aims to publish confirmed corrections within 48 hours. | |
/methodology.html and all company profiles | Clarified | The methodology defined Secondary source only in two different ways, and about 160 facts sourced to press reports carried that label. One definition now applies everywhere: Publicly disclosed means stated by a credible publication but not confirmed from a primary source; Secondary source only means the fact rests on a single aggregator or database. Facts were relabelled mechanically from their source type. Company pages no longer carry HumanAudit offers in the footer; the footer column now links to the publisher's policies. Review dates are now spread over weeks 9 to 13 after each check, within the 90-day target. | |
/use-cases/saas-integration-governance.html, /use-cases/api-key-governance.html, /use-cases/nhi-security-operations.html | Updated | Three playbooks the original plan called for were added: SaaS-to-SaaS integrations and OAuth apps, API keys and API client credentials, and NHI in security operations. Each is sourced to primary documentation and labels its OWASP NHI Top 10 mapping as ours. | |
/frameworks/eu-ai-act-nhi.html, /regulation/, /companies/category/, diagrams | Updated | An EU AI Act application timeline and a market map counting sourced profiles per category were added. Four diagrams were corrected: the identity and credential diagram called an identity the actor; the taxonomy diagram had overlapping labels; the knowledge graph diagram did not match the published data; the maturity diagram showed six levels where the model has five. | |
/news/2026/1password-acquires-apono.html, /news/2026/cisco-becomes-teleport-largest-strategic-investor.html, /news/2026/palo-alto-networks-relaunches-cyberark-as-idira.html, /news/ | Corrected | The Apono item said Swisscom Ventures confirmed its exit; its note says Apono joined 1Password, and the item now says that. The Teleport item said Teleport remains independent and venture-backed, which the announcement does not state; the sentence was removed and the related point is labelled as our reading. The Idira item and the news index said the Venafi certificate products had been renamed in March 2026; that sentence was removed. | |
/breaches/snowflake-unc5537-2024.html | Corrected | Live Nation (Ticketmaster) and Santander were listed among organisations that disclosed Snowflake-related incidents. Their own statements do not name Snowflake or the provider, and we found no primary source linking them to UNC5537, so they are no longer in that list. The page records their statements and says why. | |
/breaches/microsoft-storm-0558-2023.html | Corrected | The page presented our OWASP NHI Top 10 mapping as the root cause and paraphrased the Cyber Safety Review Board. The mapping is now labelled as our interpretation, not a finding by Microsoft, the CSRB or OWASP; statements attributed to the CSRB now quote its report; and the audit log change is dated to Microsoft's announcement of 19 July 2023. | |
/ai-agents/ai-agent-authentication.html | Corrected | The page said nested act claims carry the original principal through agent-to-agent delegation. In RFC 8693 the original principal stays in the sub claim, and section 4.1 says nested act claims are informational only and are not to be used in access control decisions. The page now says so. | |
/glossary/zero-standing-privilege.html | Corrected | Zero standing privilege was defined as issuing short-lived credentials. It concerns permissions: access is granted just in time for a task and then removed. Short-lived credentials are complementary, and a short-lived token for a role with standing permissions is not zero standing privilege. | |
/frameworks/eu-ai-act-nhi.html, /frameworks/eu-ai-act-identity-obligations.html | Corrected | The pages applied the Article 9 to 15 obligations to agentic AI generally. Those articles apply only to high-risk AI systems under Article 6; the pages now say so and label the NHI reading as ours. The Article 50 scope statement now names the system types and the exceptions. | |
/governance/index.html, /breaches/microsoft-storm-0558-2023.html, /breaches/snowflake-unc5537-2024.html, /frameworks/nis2-nhi.html, /frameworks/owasp-nhi-top-10.html, /learn/nhi-policy-template.html, /learn/non-human-identity-lifecycle.html | Corrected | After the definition correction above, these pages still called API keys, a signing key, user credentials or personal access tokens non-human identities, and the policy template said AI agents are non-human identities. They now follow the taxonomy: identities hold credentials, and an agent holds a registered non-human identity. The NHI4 and NIS2 controls now say to replace personal access tokens used by automation instead of applying MFA to them. The policy template and lifecycle guide now tie retirement to the end date of an identity's registration, not to credential expiry. | |
/companies/miniorange.html, /companies/wiz.html, /companies/compare/ (Wiz and Cortex Cloud comparisons), /companies/operant-ai.html, /companies/grip-security.html, /companies/natoma.html, /companies/keycard.html, /companies/runlayer.html, /companies/acquisition-records.html, /market/ | Corrected | miniOrange and Wiz showed no founding year although miniOrange's About page says it started in 2012 and Globes reports Wiz was founded in March 2020. Cortex Cloud comparisons showed founding and headquarters as not disclosed; they now say it is a Palo Alto Networks product announced on 13 February 2025. Company SOC 2 claims are now described as audit reports, not certifications. Cofide is described as UK-based, as Keyfactor's release says, and Cyera's release says Oasis will operate as Cyera Identity. | |
/learn/what-is-non-human-identity.html, /ai-agents/mcp-security.html | Corrected | Codecov describes the stolen credential as a Google Cloud Storage key, not a service account key. The Veza ratio was described as telemetry, which the cited page does not say, and a prior-year Palo Alto Networks ratio was not on the cited page; both were removed. A sentence on DPoP and RFC 8705 was cited to the MCP specification, which does not mention them; it now cites the RFCs. | |
/about.html, /methodology.html, /editorial-policy.html, /legal-notice.html, guide and glossary disclosure blocks | Clarified | The disclosure now says that HumanAudit Inc. is developing a paid NHI discovery and ownership product, offered now as a free guided scan, which overlaps with capabilities of companies profiled here, and that it sells ISO/IEC 42001 documentation toolkits. Oleria comparison pages, the Oleria news item and the directory axes caption now disclose Oleria's relationship with HumanAudit Inc. Negative statements about company products now say what the documentation we reviewed does not describe. | |
/tools/breach-readiness-score.html (scenario text) and /tools/owasp-nhi-top-10-audit.html | Corrected | An independent audit on 9 October 2026 found that the breach readiness tool still repeated incident details we had already corrected on the breach pages: that an Okta support engineer uploaded the HAR file (customers did), that the Snowflake accounts were service accounts (not publicly disclosed), and that Storm-0558 went undetected for months. The scenarios now match the breach pages, the CircleCI scenario says the initial access was a human engineer's session, and calling NHI the primary failure in each case is labelled as our reading. The OWASP self-audit question for NHI6 tested a different risk and was rewritten against the OWASP NHI Top 10 2025 text. Both tools no longer promise outputs they do not produce. | |
/breaches/salesloft-drift-2025.html, /breaches/ | Corrected | Google was listed as a confirmed victim of the Salesloft Drift campaign. Google said there was no compromise of Google Workspace or Alphabet itself; it is no longer listed. The other organisations named now each link to their own disclosure. | |
/companies/acquisition-records.html, /companies/directory.html, /data/companies.json | Corrected | The acquisition records page showed the Prompt Security and WideField deals as unverified although SentinelOne's filing and Cisco's update confirm completion, cited a page about Entro for WideField and an article about Permiso for Spera. The records are corrected and link to the full profiles. Oasis Security appeared twice in the directory and the data download; it now appears once, as Cyera Identity. The ownership filter placed four companies whose acquisition is only announced under part of a larger company; it now has an Acquisition announced, not completed option. | |
/learn/what-is-non-human-identity.html and other guides, /llms.txt | Corrected | Our definition pages listed API keys, tokens and certificates as types of non-human identity and gave different counts of identity types (nine on one page, eleven on another), which contradicted our own taxonomy: an identity is not a credential. The definition pages now use the taxonomy's definition and its eleven identity types, with credentials listed separately as what identities hold. Other guides that still called credentials or agents non-human identities were corrected later the same day (see the separate entry of the same date). Several pages also overstated what workload identity removes: credentials issued by SPIFFE or a managed identity can still be stolen and replayed until they expire unless they are bound to a key. | |
/companies/category/ (all category pages), /companies/oleria.html | Corrected | The category tables had a column called Ownership attribution that mixed owner assignment with acting on behalf of a user and with logging, and showed positive entries for companies whose own profile says owner assignment is not publicly verified. The three capability columns now show the status from each profile's capability table, using the labels defined in the methodology. The Oleria profile said Reviewed by the company; Oleria answered our questionnaire but, under our editorial policy, companies do not approve pages, so the field now says that. | |
Site-wide | Clarified | Pages said Last reviewed by the editor. The checks were done with AI research assistants under the editor's responsibility, so pages now say Last checked, with AI-assisted research, and name the editor. The header Book a call link no longer appears on company, comparison or category pages. Fonts are now served from this site instead of Google Fonts, and the cookie settings link now lets visitors withdraw consent. | |
/companies/sectona.html, /companies/astrix-security.html, /companies/entro-security.html, /companies/oleria.html, /companies/cyera-identity.html, /companies/microsoft-defender-for-cloud-ciem.html, /companies/saviynt.html | Corrected | An independent check of ten profiles found that Sectona's founding date, headquarters and ownership were linked to Securden's About page instead of Sectona's. The cause was in our build: two profiles that used the same internal source label shared one link, so the later profile's link replaced the earlier one. Six profiles were affected and now link to their own sources; the facts themselves did not change. The same check found that Saviynt's profile called Sachin Nayyar Chairman, which the cited release does not say, and that the Astrix profile listed Anthropic as an investor on one secondary report. Both were corrected. | |
/news/2026/eu-digital-omnibus-on-ai-enters-into-force.html, /frameworks/eu-ai-act-nhi.html, /frameworks/eu-ai-act-identity-obligations.html | Corrected | We have now read amended Article 113 of the AI Act in the consolidated text as at 27 July 2026. It confirms 2 December 2027 (Annex III) and 2 August 2028 (Annex I) for Chapter III, Sections 1 to 3, and 2 December 2026 for the two new Article 5 prohibitions. The pages said Article 50 transparency did not move; new Article 111(4) gives generative systems placed on the market before 2 August 2026 until 2 December 2026 for Article 50(2). Article 49 was not amended and is outside the deferral; how that interacts with the deferred classification rules is labelled as our interpretation. Article 4 now quotes the amended wording. This resolves the 2026-10-07 note that Article 113 had not been read word for word. | |
/companies/ (all 146 published profiles) and /companies/directory.html | Corrected | Every published company profile was checked line by line against public sources on 9 October 2026, and 1,256 changes were made, 45 of them to facts we judged materially wrong. The most serious: Natoma was described as part of Snowflake, but Snowflake has only announced an agreement to acquire it; Descope's 2025 extension was credited to a lead investor the company did not name; customer names on the Lasso Security, DigiCert, Noma Security, Oasis and Reco profiles had no company or customer source and were removed; Sectona was described as self-funded where the source says profitable; a deal value for Stytch came from an aggregator, not from Twilio; Dave McJannet was still shown as HashiCorp's CEO; JumpCloud's customer count came from an aggregator; Saviynt's 2025 funding was described as raised mainly for NHI and agent work, which its investor's release does not say; and the Astrix profile said completion was unconfirmed although Cisco confirmed it on 29 June 2026. Across the profiles, facts supported only by press are labelled Publicly disclosed and facts resting on a single aggregator Secondary source only, as the methodology defines; announced deals are no longer described as completed, and process notes were removed from public text. Six directory rows were corrected to match. | |
/breaches/okta-support-2023.html, /breaches/, /research/nhi-breach-case-library.html, /frameworks/owasp-nhi-top-10.html | Corrected | The Okta support system breach was mapped to different OWASP NHI Top 10 risks on different pages. It now has one mapping everywhere, which is ours and not OWASP's: NHI10, Human use of NHI, as the primary risk, with NHI2, Secret leakage, and NHI5, Overprivileged NHI, contributing. NHI4, Insecure authentication, was dropped because the authentication controls on the account are not publicly disclosed. | |
/research/nhi-ratio-benchmarks.html, /learn/what-is-non-human-identity.html, /learn/non-human-identity-management.html | Corrected | The 45:1 ratio was stated as a Rubrik Zero Labs finding. Rubrik's original post is no longer available and we found no other Rubrik publication that states it; the figure is now given as cited by Oasis Security and marked as not confirmed at source. Entro Labs' 144:1 figure is now cited to Entro's own release of 22 July 2025, and the Reuters reports behind the Codecov customer count and the Storm-0558 email count are now cited directly. | |
/use-cases/github-actions-aws-oidc.html, /use-cases/nhi-discovery-queries.html | Corrected | Every command and code example on the guide and playbook pages was run against the real command-line tools (AWS commands against a local AWS emulator, Terraform as OpenTofu, filters against sample data). The pipeline check on the GitHub Actions page failed on every plan, including the page's own secure configuration, and was rewritten. The workflow now uses action versions that run on Node 24, because GitHub-hosted runners stopped running Node 20 actions on 23 September 2026. The credential report example now waits for the report to finish, the GitHub inventory calls now paginate, the Azure example also lists certificates, and the CloudTrail filter now counts distinct principals. | |
Site-wide source check of guides (98 pages) | Corrected | Every hand-written guide, glossary, role, governance and research page was checked line by line against primary sources on 9 October 2026, and 1,069 corrections were applied. Most remove generalisations that had no source (for example most organisations, the most common finding), label our interpretation of laws and standards as interpretation, and repair sentences. The factual errors corrected are listed in the entries that follow. Pages changed show a review date of 9 October 2026. We also changed three statements about how this site works: guides written before October 2026 did not all go through the six-step check described on the editorial team page, the evidence labels and build checks apply to company, funding and deal records and not to every sentence in a guide, and the re-verification intervals in the methodology are targets set in October 2026 that have not yet completed a cycle. | |
/breaches/snowflake-unc5537-2024.html | Corrected | The page said the actor compromised at least 165 Snowflake customers. Mandiant said it notified approximately 165 potentially exposed organisations; the number accessed is not public. Statements that the stolen credentials were service accounts had no source and were removed, with the OWASP mapping that rested on them. Record counts for AT&T and Ticketmaster are now attributed (a company spokesperson and a seller's claim). A Santander record count, regulatory inquiries and class actions had no source and were removed. Dates for Snowflake's advisory and its multi-factor authentication changes were corrected. The page now records the guilty plea announced by the US Department of Justice on 5 August 2026. | |
/breaches/okta-support-2023.html | Corrected | The headline said the service account was on a personal laptop. Okta's root cause report says the credential was saved to an employee's personal Google profile on an Okta-managed laptop, and that compromise of that account or device was the most likely route. The page said files of every customer were exposed; files of 134 customers were accessed, 5 customers had sessions hijacked, and on 29 November 2023 Okta said a report naming all support system users had been downloaded. Six timeline dates were wrong and were corrected from the Okta, BeyondTrust, Cloudflare and 1Password accounts. Statements about missing multi-factor authentication and a shared account had no source and were removed. | |
/breaches/codecov-2021.html | Corrected | A timeline row said the FBI confirmed an investigation in May 2021; Codecov referred to a federal investigation without naming an agency. The page said tens of thousands of pipelines ran the altered script and that stolen credentials gave access to production and package registries; the number affected is not public and no disclosing company reported that. Disclosures by HashiCorp, Twilio, Rapid7 and monday.com are now given as each company stated them. Claims that the incident caused later industry initiatives were removed. | |
/breaches/shai-hulud-npm-2025.html | Corrected | The page said the worm would publish stolen secrets if maintainers cleaned up. The sources describe a different behaviour: it tries to erase the user's home directory when it can obtain neither a GitHub token nor npm credentials. The count of 25,000 repositories was attributed to Entro; it is Wiz's count of 24 November 2025, and other researchers published different totals, which the page now lists. The second wave was dated early November; the sources date it to 24 November 2025. Entro's count of 1,195 organisations is now cited through a secondary report because Entro's page no longer resolves. | |
/frameworks/csa-agentic-trust-framework.html, /frameworks/standards-tracker.html, /ai-agents/agent-registries.html | Corrected | We described the Agentic Trust Framework as a Cloud Security Alliance framework and named two authors. It is an open specification by Josh Woodruff of MassiveScale.AI, introduced on the CSA blog on 2 February 2026 and still a public review draft. We also said that CSA's MAESTRO, the IETF agent draft and a NIST concept paper all prescribe agent registries, cryptographic identities and short-lived credentials. They do not: MAESTRO lists agent impersonation as a threat with other mitigations, and the NIST paper does not mention registries. | |
/frameworks/dora-nhi.html, /frameworks/nis2-nhi.html, /learn/nhi-financial-services.html | Corrected | Article references were wrong. In DORA, Article 9(4) point (c) covers access rights, (d) authentication and cryptographic keys, and (e) change management; a major ICT-related incident is defined in Article 3(10); information sharing is Article 45 only; and only major incidents are reportable. In NIS2, incident reporting is Article 23, not Articles 23 and 24. | |
/learn/post-quantum-machine-identity.html, /learn/certificate-lifecycle-management.html | Corrected | We said NIST deprecates RSA-2048 and P-256 after 2030. NIST IR 8547 is still an initial public draft; it would deprecate 112-bit strength algorithms after 2030 and disallow the rest after 2035, and P-256 is not 112-bit. A sentence about a US executive order setting 2035 was wrong: Executive Order 14412 of 22 June 2026 sets 2030 and 2031 dates for certain federal systems. We attributed a post-quantum rationale and a statement about devices to the CA/Browser Forum ballot on certificate lifetimes; neither is in the ballot. | |
/use-cases/github-actions-aws-oidc.html, /use-cases/nhi-discovery-queries.html | Corrected | The guide said every GitHub OIDC token carries the audience sts.amazonaws.com; the default audience is the repository owner's URL and the AWS action requests that value. A CI check script passed policies it should have failed and was fixed. The subject format for repositories created after 15 July 2026 was added. An AWS CLI example used an invalid page size and a jq filter that did not run; both were fixed. | |
/ai-agents/mcp-security.html, /ai-agents/mcp-server-governance.html, /frameworks/oauth-for-nhi.html, /ai-agents/what-is-ai-agent-identity.html | Corrected | Three more pages said the Model Context Protocol mandates OAuth 2.1; authorization is optional in the specification. OAuth 2.1 was described as a finished standard; it is an IETF draft. The act claim in RFC 8693 was described as naming the principal; it names the acting party. References to an individual IETF agent draft were updated to the working group draft that replaced it. | |
/research/nhi-ratio-benchmarks.html, /learn/what-is-non-human-identity.html, /corrections.html (entry of 2026-09-09) | Corrected | An earlier entry in this log said the Cloud Security Alliance's 2024 survey states a ratio of 20 to 1. It does not. The figure appears once, as background, in the press release for the survey, where it links to the vendor that commissioned it. It is no longer presented as a survey result. Citations of Entro's 2025 report now point to press coverage because Entro's page no longer resolves. | |
/frameworks/standards-tracker.html, /regulation/ | Updated | All remaining tracker rows were opened on 9 October 2026. Changes: ISO/IEC 27001:2022 has Amendment 1:2024; NIST states that AI RMF 1.0 is being revised; OAuth 2.1 was added as a draft (version 16); the NIST agent identity project published a summary of comments on 29 September 2026; ISO/IEC 27090 is at the publication stage and not yet published. | |
/roles/, /frameworks/ (guide pages) | Improved | Second promotional boxes for HumanAudit products were removed from guide pages. Each guide now carries one closing offer, as the box itself says. | |
/research/nhi-ratio-benchmarks.html#counting | Withdrawn | We withdrew the planned 2027 NHI Management Benchmark survey before any data was collected. The page committed us to a minimum of 120 qualifying responses, and we do not have a respondent base that makes that commitment credible. No results existed, so none are affected. The counting definitions drafted for it are kept on the ratio benchmarks page, where the old address now redirects. | |
/frameworks/standards-tracker.html, /frameworks/wimse.html, /ai-agents/agent-identity-standards.html, /frameworks/oauth-for-nhi.html | Updated | The standards tracker carried a status date of 5 August 2026. Its IETF and OpenID rows were checked again on 7 October 2026 and every row now shows the date its source was last opened. Changes: the individual draft draft-klrc-aiagent-auth was adopted by the WIMSE working group and replaced by draft-ietf-wimse-aims (15 September 2026); WIMSE draft versions were updated and one informational document has been submitted to the IESG; the AuthZEN Authorization API 1.0 is a Final Specification (12 January 2026); Transaction Tokens is at draft 11. Regulation rows are now generated from the regulation tracker. We previously wrote that the Model Context Protocol mandates OAuth 2.1; the specification makes authorization optional and bases it on OAuth 2.1 where it is used. Unsourced forecast dates for WIMSE publication were replaced by a labelled estimate. | |
/frameworks/csa-nhi-program-guide.html | Corrected | The page attributed to the Cloud Security Alliance survey figures that are not in it: incidents at around two-thirds of organisations, inventory completeness below 50%, and a ratio rising year on year. These were removed. The page now gives the findings on CSA's report page (818 respondents, published 11 September 2024, commissioned by Astrix Security) and says the survey was a single report, not an annual series. | |
/breaches/microsoft-storm-0558-2023.html | Corrected | The page said CISA issued Emergency Directive 23-02 on this incident. We found no such directive. CISA and the FBI published joint advisory AA23-193A on 12 July 2023, and the page now cites it. The figure of about 60,000 State Department emails is now attributed to press reports. | |
/ai-agents/multi-agent-authorization.html, /breaches/salesloft-drift-2025.html, /breaches/shai-hulud-npm-2025.html, /breaches/snowflake-unc5537-2024.html, /breaches/codecov-2021.html, /learn/ (four guides) | Corrected | Figures without a source were sourced or removed. Agent2Agent adoption figures are now attributed to the Linux Foundation announcement of 9 April 2026, and a version 1.0.1 release we could not confirm was removed. The figure of more than 700 organisations in the Salesloft Drift incident is attributed to Google and described as potential exposure. Statements about regulatory inquiries and notifications, class actions, a count of 29,000 Codecov pipelines, an illustrative historical ratio of 10:1 and numeric thresholds presented as norms were removed or labelled as our own suggestions. | |
/breaches/salesloft-drift-2025.html | Corrected | The page said customers' security teams, including Cloudflare's, detected the activity themselves. Cloudflare's own account says Salesforce and Salesloft notified it on 23 August 2025. The detection section, timeline and attack chain were rewritten from Google Threat Intelligence Group, Cloudflare and the findings Salesloft published with Mandiant: the earlier access to a Salesloft GitHub account, the 8 to 18 August activity window and the token revocation on 20 August. Summaries that gave 700 organisations as a count of breaches now give it as Google's estimate of potential exposure. Unsourced statements about scope, market effects and customer review practices were removed. | |
/breaches/microsoft-storm-0558-2023.html (key theft) | Corrected | The page presented Microsoft's September 2023 crash dump explanation as fact. Microsoft corrected that explanation on 12 March 2024, and the Cyber Safety Review Board found that Microsoft does not know how or when the key was stolen. The summary, timeline, attack chain and lessons now say so. | |
/frameworks/csa-nhi-program-guide.html (scope) | Corrected | The page was titled as a programme guide and listed recommendations as CSA's. The source is a single survey report. The page was retitled, the recommendations are now labelled as our own reading, and a ratio of 20 to 1 is no longer presented there as a survey result because we could not confirm it from the report page. | |
/frameworks/iso-27001-machine-identity.html, /frameworks/soc2-nhi.html | Improved | Each page carried two overlapping mapping tables after the site merger. Each now has one. The column that paraphrased control and criteria wording was replaced with short summaries in our own words. | |
/frameworks/iso-42001-nhi.html | Updated | Annex A objective numbers, titles and control counts (38 controls under nine objectives, A.2 to A.10) were checked against two published listings. Two objective titles were corrected to the wording of the standard: A.2 is Policies related to AI, and A.3 is Internal organization. The note saying the numbering was unconfirmed was replaced with a statement of what was checked and against what. | |
22 additional company records | Updated | Records were researched and published for 22 tracked companies that previously had only a directory line, including Prompt Security (SentinelOne), Protect AI (Palo Alto Networks), Spera Security and Axiom Security (Okta), Fabrix Security (Silverfort), Anetac (Quest Software), WideField (Cisco), EmpowerID (Omada), Cofide, Eos, Aim Security, Neo, AIR, Aizome, MintMCP, Opnova, Zaperon, Sectona, miniOrange, Encryption Consulting, Clarity Security and Tumeryk. One company, ARCON, remains a directory entry because its ownership could not be sourced. Earlier unverified notes were corrected: Opnova is an identity governance product for applications without an API, not an NHI-native platform; the Okta and Spera agreement dates from December 2023. | |
/regulation/, /news/ (EU Digital Omnibus on AI) | Updated | The entry into force of Regulation (EU) 2026/1744 (27 July 2026) and its publication (24 July 2026) are now sourced to EUR-Lex. The two application dates for high-risk systems are confirmed from recital 40; the amended Article 113 has not yet been read word for word. | |
Older guides (site-wide) | Corrected | Unsourced statistics and ranking language were removed or reworded on older pages, including claims that discovery finds two to five times the tracked count, that stored keys are a recurring cause of publicly reported cloud breaches, and that named risks appear in most breaches. Glossary entries for NIS2 and DORA now say that applying them to non-human identities is our interpretation. Company profiles no longer carry a page-level Verified badge; each fact carries its own label. | |
New models and open data | Improved | Added the NHI taxonomy, the human accountability chain, the governance canvas and the risk model as versioned NHI Governance models, and published the claims and evidence dataset, the knowledge graph, an acquisition table and an investor table. | |
124 company profiles (second audit pass) | Updated | The second-pass source audit covered the 124 company profiles published at that point. Across those records, 783 facts were confirmed, 511 corrected, 230 added, 48 given a weaker label and 38 removed or marked unsupported. Material corrections include: Wiz is part of Google (completed 11 March 2026); StrongDM is part of Delinea (completed 5 March 2026); ConductorOne renamed itself C1 on 6 April 2026; the Venafi name: retirement not announced; venafi.com redirects to Palo Alto Networks Next-Generation Trust Security; ServiceNow's filing gives the Veza price as about $1.2 billion; Sysdig's total funding is $744M, not $394M; SSH Communications Security's chief executive is Lars Bell from 1 October 2026; Doppler's chief executive is Amber Britton; Omada's Agent Governance is announced, not launched; AppOmni's headquarters is San Mateo and its chief executive is Neill Occhiogrosso; the Linx Security chief executive's background had been confused with a co-founder's. Each profile lists what could not be verified. | |
/learn/nhi-glossary.html | Corrected | The glossary described Astrix Security as an early-stage vendor and carried out-of-date entries for Entro Security, Oasis Security and Venafi. Company entries now state current ownership from the audited records. Twenty-five terms were merged from the nhimanagement.com glossary, and the list markup was made valid. | |
Company pages (site structure) | Improved | Company pages moved from /vendors/ to /companies/. Every old address redirects permanently to the new one in a single step. The Cyera Identity record is now at /companies/cyera-identity.html. | |
nhimanagement.com merge | Improved | The remaining held nhimanagement.com pages were merged into their counterparts here: the ISO/IEC 42001, ISO/IEC 27001, SOC 2 and OWASP guides, discovery, lifecycle, secrets and rotation guides, the glossary, the corrections log, the reading paths and the lifecycle readiness check. All page-level redirects from nhimanagement.com are now released. | |
/news/, /companies/compare/, /editorial-team.html | Improved | Added a newsroom with sourced analysis, side by side comparisons for companies that our records list as direct competitors of each other, and a page naming the accountable editor. | |
30 company profiles (deep research pass) | Updated | Thirty profiles were re-researched against primary sources and extended with leadership, products, a 21-row capability matrix, customers, competitors and recent news. Material corrections: Strata Identity is now part of Rubrik (announced 9 June 2026) and was listed as independent; ServiceNow's acquisition of Veza closed on 2 March 2026; Palo Alto Networks relaunched the CyberArk platform as Idira on 12 May 2026; Keyfactor's acquisition of Cofide is still an announced intent, not a completed deal; Britive's Series B was in March 2023, not 2022; NewCore's funding was announced on 15 June 2026 and its reported valuation is press-only; an unsupported pre-seed round was removed from Unosecur; KuppingerCole placements for Andromeda Security and Britive were reworded to match the public report page; and estimated funding rows for Aembit, Clutch Security and P0 Security were replaced with the announced rounds. | |
Company profiles (110 pages) | Corrected | An audit on 6 October 2026 found that evidence labels did not follow our own methodology: 325 facts were labelled Verified on a single press or aggregator source, and 192 facts labelled Publicly disclosed rested on a single aggregator. Those labels were downgraded to Publicly disclosed or to a new label, Secondary source only. No fact was changed by this correction; only the label describing how well it is supported. | |
/companies/, /companies/compare/, /companies/rfp-questions.html, /roles/procurement.html, /market/ | Corrected | These pages still described Cyera's acquisition of Oasis Security as a letter of intent. Cyera announced completion on 3 September 2026. The vendor index also called Oasis's March 2026 round a Series C with a total of $190M+; the company's release says Series B and $195M. | |
/companies/astrix-security.html, /companies/entro-security.html, /companies/cyera-identity.html, /companies/oleria.html, /companies/token-security.html | Updated | Re-sourced to primary announcements. Astrix: Cisco confirmed completion on 29 June 2026 and has not disclosed a price. Entro: the June 2024 Series A was led by Dell Technologies Capital; SailPoint states that terms of the acquisition were not disclosed. Cyera Identity: completion dated 3 September 2026 and Oasis's funding history restored. Oleria: chief executive change of 1 September 2026 sourced to the company's release. | |
/companies/cyera-identity.html | Corrected | The profile still described the Cyera acquisition as a pending letter of intent after the tracker had been corrected on 9 September. Ownership, title and buyer guidance now reflect the completed acquisition (Oasis operates as Cyera Identity). | |
/ (homepage) | Corrected | The homepage ratio counter rendered as 0:1 without JavaScript and implied a single 45:1 figure. It now states the published range (17:1 to 144:1) and links the explanation of why the figures differ. | |
/learn/non-human-identity-security.html | Improved | Merged into /learn/non-human-identity-security.html. The two pages answered the same question; the condensed primer now redirects to the complete one. | |
Site-wide | Improved | nhimanagement.com content is being merged into this site. Migrated pages carry a 'Migrated, under review' label until re-verified. See /methodology.html. | |
/market/nhi-consolidation-tracker.html | Corrected | The tracker recorded the Cyera acquisition of Oasis Security as a letter of intent that had not closed. Cyera completed the acquisition on 3 September 2026 at the reported $1B. Buyer guidance that told readers to treat the Oasis position as uncertain has been replaced. Our fortnightly re-verification cycle did not run, which is how a closed transaction stayed marked as pending for six days. | |
/market/nhi-consolidation-tracker.html | Updated | Three acquisitions inside our own stated May to July 2026 window were missing: Snowflake and Natoma (announced 27 May 2026), Okta and Permiso Security (announced 30 July 2026, ~$200M reported, completed 26 August 2026), and Zscaler and Symmetry Systems (21 May 2026). The combined reported value has been revised from $1.6bn to $1.9bn, or approximately $2.2bn including 1Password and Apono. The claim that three leading platforms were acquired has been revised to five companies. | |
Site-wide | Corrected | Five pages stated that non-human identities became "the leading root cause of enterprise breaches". This is not supportable. Verizon's 2026 DBIR found vulnerability exploitation, not credential abuse, was the leading initial access vector, and credential abuse as an entry point fell from 22% to 13%. The claim has been replaced with the defensible version on every page, and the DBIR figures are now cited on /about. | |
/learn/nhi-glossary.html | Corrected | The Oasis Security Series B was described as Sequoia-led. It was led by Craft Ventures, with Sequoia, Cyberstarts and Accel participating. Total funding is $195M, not "$190M+". | |
/breaches/microsoft-storm-0558-2023.html | Corrected | The page used Microsoft's initial July 2023 estimate of approximately 25 affected organisations. That figure was superseded by the Cyber Safety Review Board report of 2 April 2024, which established 22 organisations and 503 individuals. The CSRB report is now cited as a primary source. | |
Site-wide | Corrected | Five pages described the Shai-Hulud 2.0 worm as compromising approximately 25,000 repositories. Those were attacker-created repositories used to dump stolen secrets, not compromised victim repositories. The wording now says "attacker-created exfiltration repositories". The 1,195 affected organisations figure from Entro Security is unchanged and correct. | |
Site-wide | Corrected | Two pages attributed an NHI-to-human ratio range of "20:1 to 50:1" to the Cloud Security Alliance. CSA's 2024 survey states a factor of 20 to 1 only. Higher figures in circulation are vendor-published. The attribution has been corrected and the provenance made explicit. | |
/index.html | Corrected | The homepage stated a single ratio of 45 to 1 as fact in the hero and sourced it in the stat block to Rubrik Zero Labs, a vendor. Our editorial policy prefers non-vendor sources over vendor claims, and our own definitional page already treats the ratio as disputed. The hero no longer states a figure and the stat now shows the 20:1 to 45:1 range with its provenance. | |
/index.html | Updated | The homepage AI-agent incident statistic now discloses that the CSA report it comes from was sponsored by Token Security, a vendor in this market, and names the study and sample size. We already applied this standard to the Strata Identity-sponsored survey; it should have been applied here from the start. | |
Site-wide | Updated | Added linked primary sources to all six breach post-mortems and to five framework and market pages. Our editorial policy requires sources to be "named, verifiable, and accessible", and our about page says pages are written against primary sources. Sources were named in prose but almost never linked: across the site there were 20 outbound links to primary sources against 837 to our own properties. Naming Mandiant, CISA and the CSRB without linking them did not meet the standard we published. | |
Site-wide | Updated | Hedged three figures that were stated more firmly than the evidence supports: Ticketmaster's 560M records is a claim made by the sellers and was not confirmed by Live Nation; AT&T's ~$370K ransom was reported by Wired and has never been confirmed by AT&T; and Okta's 18,400 is the company's total customer count used by the press to size the revision, not a victim figure Okta published. The CSA audit statistic now says respondents "doubted they could pass" rather than "could not pass". | |
Site-wide | Improved | Fixed a grid rule that pushed the layout 64px wider than the viewport at 280px, and raised header, footer and navigation tap targets to 44px on touch devices to meet WCAG 2.2 AA target size. Inline links inside sentences are exempt and were left alone. | |
Site-wide | Improved | "Last reviewed" now reflects genuine per-page review rather than a single build date. Only the 20 pages actually re-verified on 9 September 2026 carry that date; the rest still show 5 August 2026, which is the truth. | |
/frameworks/eu-ai-act-nhi.html | Corrected | Pages described the Digital Omnibus by its legislative milestones without naming the instrument. It is Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force from 27 July 2026. The regulation is now named wherever it is referenced. | |
/frameworks/eu-ai-act-identity-obligations.html | Corrected | Pages stated the deferred high-risk dates without stating what did not defer. Article 50 transparency, Article 49 registration and national market surveillance enforcement powers apply from 2 August 2026, and Article 4 AI literacy has applied since 2 February 2025 at every risk tier. A live-obligations block has been added to both EU AI Act pages. | |
/frameworks/iso-42001-nhi.html | Updated | Added the specific reason ISO/IEC 42001 does not confer EU AI Act conformity: it is not a harmonised standard and gives no Article 40 presumption of conformity. CEN-CENELEC JTC 21 assessed it against the Article 17 quality management requirement, found it did not align, and drafted prEN 18286, whose annex maps onto ISO/IEC 42001 Annex A controls. | |
Site-wide | Improved | Added cross-references to the sister publications from HumanAudit Inc. in the footer, in llms.txt and contextually on eight framework and role pages. Publisher structured data now links the four properties through sameAs. | |
Site-wide | Improved | House style: long dashes removed from all content and normalised on every build. Duplicate Organization structured data on the homepage removed. | |
/frameworks/csa-agentic-trust-framework.html | Corrected | Our own error, found in internal validation. The page stated the Agentic Trust Framework was co-authored by John Kindervag. It was authored by Josh Woodruff (MassiveScale.AI) and Michelle Savage; Kindervag wrote the foreword to the companion book. The page also described a five-stage maturity progression that we had inferred rather than verified. The framework in fact defines five core elements and four maturity levels named Intern through Principal. Both corrected against the CSA publication and the published specification. | |
/frameworks/csa-agentic-trust-framework.html | Updated | Added disclosure that the ATF authoring consultancy offers paid implementation and certification against the framework, and that the companion 285-respondent survey was published jointly with Strata Identity, a vendor in this market. | |
/frameworks/eu-ai-act-nhi.html | Corrected | Page presented 2 August 2026 as the live compliance date for high-risk obligations. The Digital Omnibus on AI was endorsed by the European Parliament on 16 June 2026 and approved by the Council on 29 June 2026, deferring Annex III stand-alone high-risk obligations to 2 December 2027 and Annex I to 2 August 2028. Timeline rewritten; Article 50(2) and new Article 5 prohibition dates added. | |
/frameworks/wimse.html | Updated | WIMSE has restructured from a single service-to-service protocol draft into a document family. Current drafts listed with status: architecture, workload identifier, workload credentials, workload proof token, and workload identity practices. | |
/ai-agents/agent-identity-standards.html | Updated | NIST NCCoE concept paper detail verified against the primary publication (5 February 2026, comment closed 2 April 2026). Added NIST COSAiS and the OWASP Top 10 for Agentic Applications. | |
Site-wide | Improved | Repaired structurally invalid HTML on 39 pages (unclosed main and article elements, orphaned footer tags, unbalanced containers) and restored the missing footer navigation block on 30 pages. | |
Site-wide | Improved | Authorship standardised to the HumanAudit Inc. editorial team across all pages, with a corresponding organisation entity in structured data. | |
/companies/astrix-security.html | Corrected | Page described the Cisco acquisition as unconfirmed reporting at $250M to $350M. Cisco announced intent to acquire on 4 May 2026 and has since reported completion; Calcalist reported approximately $400M. Ownership row, lede, and analysis section rewritten. | |
/companies/entro-security.html | Corrected | Page presented Entro as an independent vendor. SailPoint completed its acquisition of Entro on 29 June 2026 at a reported ~$200M. Ownership row and status banner added. | |
/companies/cyera-identity.html | Updated | Cyera signed a letter of intent to acquire Oasis on 28 July 2026 for a reported ~$1B. Recorded as a signed LOI, not a closed transaction. Ownership row and open-questions section updated. | |
/companies/ | Updated | Added a category-consolidation summary covering the Astrix, Entro, and Oasis transactions between May and July 2026. | |
Site-wide | Improved | Added BreadcrumbList structured data to 49 pages, FAQPage structured data where FAQ sections exist, and corrected 38 over-length page titles and 43 over-length meta descriptions. |
Entries carried over from nhimanagement.com
These 9 entries were logged on nhimanagement.com before the two sites were combined. They are kept here unchanged in substance so that the record stays complete. Page links point to the page that now holds the content.
| Date | Page | Type | What changed |
|---|---|---|---|
| 6 Aug 2026 | NHI ratio benchmarks | Carried over | Methodology misclassification. The survey-versus-telemetry section, added earlier the same day, described Veza’s 17:1 as survey-derived. It is not: Veza’s 2026 State of Identity & Access Report (11 Dec 2025) is built from proprietary telemetry across millions of identities and billions of permissions. Corrected, the figure is now verified and linked, and the section now notes that the two telemetry-derived figures are the highest and lowest in the table, so methodology class alone does not predict the result. |
| 6 Aug 2026 | Site-wide, Privacy policy, Contact | Carried over | Data collection removed. The tool-report email capture, its lead store and the transactional mail integration have been deleted entirely, handler, mail library and stored records. The privacy policy previously described collecting and retaining email addresses for tool reports; that description was accurate when written and is now obsolete, so the policy has been rewritten rather than amended. This site now processes no personal data beyond analytics you can decline. Anyone who submitted an address before this date can request confirmation of erasure at hello@nhigovernance.com. |
| 6 Aug 2026 | Site-wide | Carried over | Authorship disclosure. Article schema previously named the site itself as author, which told a reader nothing about accountability. Pages now carry an explicit organisational byline (HumanAudit Inc. Editorial Team) in both the visible metadata and the structured data, linked to a new editorial team page that sets out competencies and the review process. That page also states plainly that an organisational byline is a weaker trust signal than a named author with verifiable credentials, and that readers should weigh the cited sources and the correction record instead. Note added 9 October 2026: since 9 October 2026 the structured data names HumanAudit Inc. as author and Akshay Dubey as editor, matching the visible byline. |
| 6 Aug 2026 | NHI ratio benchmarks, What is NHI management | Carried over | Misattribution. Both pages credited the 82:1 figure to Rubrik Zero Labs as its own research finding. Rubrik’s Identity Crisis report presents 82:1 as an external figure (“by some measures”) with a footnote, rather than originating it; the figure comes from the Identity Security Landscape survey series (CyberArk, now Palo Alto Networks), which reports 109:1 for 2026. Restated on both pages, and the benchmark table now separates the originating source from the onward citation. |
| 6 Aug 2026 | NHI ratio benchmarks | Carried over | Incomplete methodology. The page listed four reasons published ratios diverge and omitted the most consequential one: some figures are practitioner surveys and others are product telemetry. Surveys are structurally biased low because they cannot count identities the respondent does not know exist. Added as a fifth reason. Also corrected the Entro Labs row, which stated a cloud-native scope that Entro’s own release does not claim, the scope discrepancy between the publisher and its secondary coverage is now shown rather than resolved silently. |
| 6 Aug 2026 | NHI ratio benchmarks | Carried over | Source verification. The Palo Alto Networks and Entro Labs figures were marked pending verification at publication. Both have been confirmed against the publishers’ own pages and are now linked and dated. The Veza 17:1 figure could not be traced to a primary report and is now marked unverified in both the table and the sources block. It is retained because removing the lowest published figure would bias the stated range upward. |
| 6 Aug 2026 | Homepage, What is NHI management, Least privilege | Carried over | Stale figure. These pages led with a single non-human to human identity ratio of 45:1 (Rubrik Zero Labs). The attribution was correct, but the figure had been superseded by later published research, notably 109:1 in Palo Alto Networks' 2026 Identity Security Landscape (n=2,930), up from 82:1 the prior year. Presenting one figure as settled also misrepresented a genuinely contested measurement. Replaced with a sourced range and a new benchmark comparison page setting out every published figure and why they diverge. |
| 6 Aug 2026 | Homepage | Carried over | Imprecise date. The Palo Alto Networks acquisition of CyberArk was described as a 2026 event without distinguishing announcement from completion. The agreement was announced 30 July 2025 and the transaction completed 11 February 2026. Corrected. |
| 6 Aug 2026 | Methodology | Carried over | Incomplete disclosure. The methodology page stated that we do not accept payment for favourable coverage, but did not state plainly that HumanAudit Inc. sells compliance toolkits and advisory services and that this site is a route to them. Rewritten to disclose the commercial relationship in full, and to set out where commercial links appear and where they deliberately do not. |
Found an error?
Corrections are welcome from anyone, including the vendors we write about. We evaluate every report against primary sources and publish the outcome here whether or not we agree with the reporter. Send us the detail with the form below, or email hello@nhigovernance.com.