Full disclosure of which AI tools AI Leakage uses, how we use them, and the limits we apply.

The short version

AI Leakage uses AI for research, drafting and checking, and much of the work on this site is done with it. The primary tool is Claude, made by Anthropic. What we promise is this:

Every rating rests on the vendor’s own published documents. Those documents are monitored daily. Changes that touch a claim a rating rests on are assessed and dated, and the assessment is published. When we get something wrong, we correct it and say so.

We do not claim that a person checks every claim, and we do not claim the work is fully automated. Both would be untrue. Updated 25 October 2026: this section previously said every claim in every profile is checked by a person and every cited source verified by a human. That overstated it, and our public promise since 31 August has been the one above. See Corrections.

After a profile is live, the vendor’s documents are monitored daily, and that part is automated. A change that touches a claim a rating rests on is read by a person. Routine changes — navigation, formatting, a page reissued without altering what it says, or a change the vendor reverses days later — are assessed automatically, dated and recorded, not read by anyone. Where a rating is recalculated rather than re-read, because the scoring method changed rather than the vendor did, the profile records what caused it.

We would rather tell you where the human attention actually goes than imply it is everywhere. A claim that a person reads every one of the thousands of changes these vendors make would not be true, and we would rather you could check what we say than be reassured by it.

Primary tool: Claude (Anthropic)

We use Claude to read vendors’ documents and other sources, check our claims against them, draft and edit pages, and apply our published scoring method. We no longer describe the plan or contract terms we use it under; an earlier version of this page did, and we could not stand behind the description.

Claude is also rated on this site. We score it from Anthropic’s own published terms, using the same method, the same anchors and the same evidence standard as every other vendor, and we tell you so you can weigh that. Where a vendor we rate names Anthropic as one of its providers, we say so in what we send them. Updated 25 October 2026: this page previously said the Claude profile applies a deliberate counter-correction to avoid favourable bias. We withdrew that on 28 September (see the methodology): deliberately scoring one vendor harder than the rest is a different bias, not the absence of one.

Other AI tools

We occasionally cross-check a fact with another AI tool. Any claim we publish cites the vendor’s own document, not the tool. Updated 25 October 2026: this section previously named the plans we used other tools on, and said no other AI tools were used. We could not stand behind either statement, so we removed them.

Where a person decides

• Scores. Scores come from applying our published method to the vendor’s own documents, and AI does much of that work; the method is what makes it checkable. Where a score turns on court pleadings about a named company rather than on the vendor’s own documents, a named person approves it before it is published.
• Rating changes, legal questions and anything published in the operator’s name are decided by the operator, Quentin Donald.
• “Last verified” on a profile is the date the vendor’s primary sources were last read end to end. The daily monitoring does not change it.

If you have concerns about our AI use

Reasonable people can disagree about whether a site critiquing AI vendors should use AI tools in its workflow. Our position: refusing to use AI for research would not produce better content, and pretending we do not use it would be dishonest. The right response is transparent disclosure, a published method anyone can check against the vendor’s own documents, published corrections, and the same standard for our own AI vendor as for every other.

If you spot a claim on this site that appears to be an AI hallucination, a fabricated CVE number, an invented incident, a misquoted source, please email the operator via the contact page. We will verify, correct, and publish a correction note.