How EAU researches vendor AI features

EAU focuses on AI features deployed by enterprise software vendors in organisational contexts — specifically features that may constitute high-risk AI systems under Annex III of the EU AI Act. Current coverage prioritises workforce and HR AI categories, which are most immediately relevant to enterprise deployers. Full Annex III coverage is in progress.

For each feature we record:

• Default state — whether the feature is on, off, or opt-in when a customer first enables the product
• Risk classification — our independent assessment against the applicable Annex III category
• Obligation gap — the specific deployer obligations triggered if the feature is active
• Admin path — where in the vendor admin console the feature can be reviewed or disabled (where verified from public documentation)

EAU only surfaces what is verifiable from public sources. When we cannot find public documentation for a feature behaviour, we record a documentation gap — not an assertion that the documentation does not exist. Vendors routinely maintain internal documentation, support articles, and contractual commitments that we cannot access.

A documentation gap finding means: this information is not accessible from publicly available sources at the time of our last assessment. It is not a statement about what the vendor does or does not do internally.

Vendors who believe a gap finding is addressable by public documentation are encouraged to submit a correction. See the correction process below.

Every finding is grounded in at least one source, tiered by reliability. The tier is visible on every source record so you can calibrate how much weight to place on each finding.

Each finding carries a confidence level that reflects the strength and recency of our evidence.

EAU is deliberately scoped to what public research can reliably support. We do not:

• Determine whether your organisation is compliant with the EU AI Act or any other regulation. That determination requires legal expertise and context about your specific deployment that EAU does not have.
• Score vendors as compliant or non-compliant. EAU surfaces feature behaviour and obligation exposure — not a verdict. Two organisations deploying the same vendor feature may have different compliance positions.
• Provide legal advice. Nothing on this site is legal advice. Consult qualified legal counsel for compliance decisions.
• Access vendor admin consoles, internal documentation, or non-public agreements. Our research is bounded by what is publicly accessible.
• Cover standalone consumer AI tools, AI infrastructure providers, or model APIs. EAU focuses on enterprise software vendors deploying AI features in customer environments.

For each finding, we work through this hierarchy and stop at the first method that returns meaningful content. The method used is recorded on every source record.

AI vendor features change frequently. EAU monitors publicly accessible vendor documentation for changes. Every finding is stamped with an “EAU last assessed” date reflecting when a researcher last reviewed the relevant sources.

Findings are automatically flagged as stale when the last verified date exceeds 90 days. Stale findings remain published but are visually marked — treat them with additional caution until re-verified.

EAU research may lag vendor product changes. When a vendor updates a feature, there is typically a gap between the change and our re-assessment. We disclose our last assessed date precisely so you can account for this.

If you are a vendor and believe a finding is inaccurate or incomplete, EAU welcomes corrections. The process is the same regardless of whether you are a vendor, analyst, or any other party.

1. Submit a correction → with the finding name, the specific claim you believe is incorrect, and a link to your public documentation that addresses it.
2. We will acknowledge within 2 business days.
3. We will review and publish a correction or reasoned rebuttal within 5 business days of receiving sufficient evidence.
4. All corrections are logged in the finding’s audit record with the original classification preserved.

Vendor submissions do not receive expedited review. EAU does not accept payment from vendors for classification, editorial review, or preferential placement. No vendor has access to classifications before publication.

EAU is funded entirely by subscription revenue from deploying organisations — the people whose interests are served by accurate, independent intelligence. No vendor relationship influences classification.