About

AI Security Tracker explains AI security risks in plain language for security leaders. Each entry covers what was publicly reported, why it matters, what a board may ask, and which controls help.

How an entry gets published

Editors work from public sources. Every factual claim needs a cited source with a publisher, a link, and a date. Every entry is reviewed before it is published.

AI Security Tracker publishes under its own name. Coverage is attributed to the editors of AI Security Tracker.

How the site is funded

The site is free to read and is funded by sponsors. Sponsors pay for a placement that is labeled Sponsored. They do not choose, edit, review, or veto coverage, and a sponsor is not placed next to an entry about its own product category.

Editorial standards

1. Patterns, not verdicts. We describe what was publicly reported, with attribution (“X reported…”, “according to…”). We do not state or imply that a named company was negligent, at fault, or in breach of a duty.
2. Every factual claim is cited. Each risk entry and regulation item links to at least one public source with a date.
3. Opinion and sponsorship are labeled. Analysis is marked “Analysis / Opinion”. Paid placements are marked “Sponsored” on the site and in email.
4. Conflicts of interest. We do not cover organizations where the editors have a business or professional relationship that could affect, or appear to affect, our coverage.
5. No leaked, stolen, or confidential material. We don’t quote, link, or summarize it, and we don’t link to leak sites.
6. No personal data about individuals, including health information.
7. Corrections are visible. We show a dated note on the entry and in its change log. We never quietly rewrite facts. See Corrections.

How we rate severity

Each risk is rated High, Elevated, or Watch. We score three things from 1 to 3.

• Exploitation: Has it happened in real life, been shown by researchers, or is it only theory?
• Impact: How much harm could it do, from minor to severe?
• Reach: Does it affect tools most organizations use, a large group, or a niche?

We add the scores. A total of 5 or less is Watch, and 6 or more is Elevated. A risk is rated High only if it scores 8 or 9, has happened in real life, and could lead to system takeover, stolen keys, or financial loss, could put people’s safety at risk, or has exposed more than 100,000 people’s records or compromised that many users. Ratings are our editors’ judgment based on cited public reports, not a verdict on any company. When new reports arrive, we re-score and note the change.

AI Security Tracker is general information, not legal, security, or financial advice. See the Disclaimer.

Questions: [email protected]