Editorial standards & methodology
Sourcing
- Every factual claim links to a primary source: judgments and orders on official court systems (BAILII, CourtListener/PACER, CanLII, AustLII, SAFLII), regulator publications, or official legislative texts. Press coverage supplements; it never carries a claim alone.
- Quotes are verbatim from the primary document, with pin citations. We never reconstruct, paraphrase-as-quote, or let an AI model produce a quotation.
- Numbers are datestamped ("as of"), traced to the original dataset.
Tracker methodology
- Inclusion rule: an incident enters the tracker when a court, tribunal, or regulator has made a finding, order, or public statement about AI misuse in a legal proceeding. We record their findings; we don't make our own.
- Every entry is verified against the primary record before inclusion. Entries we cannot yet verify are labelledunverified — visibly.
- We track the regulatory aftermath — referrals to and dispositions by bodies like the SRA, SDT, BSB, law societies, and state bars — and update entries as outcomes land.
- The tracker is not complete and doesn't claim to be: it contains every incident we have verified. Corrections and submissions are welcome and credited.
- The dataset is free to use under CC BY 4.0: download JSON.
Tool reviews
- Scorecards assess only verifiable, public criteria: contractual terms, security documentation, sub-processor lists, data residency options, published accuracy evidence and independent benchmarks.
- Vendors receive our findings and at least five business days to respond before publication.
- LegalAI Space, our owner's product, is never scored on this site.
- "Not published" is reported as a finding, not silently skipped.
How we use AI
This publication is produced with substantial AI assistance — research gathering, drafting, data processing, and illustration. Every piece then passes an adversarial fact-check against primary sources and a named human editor's review; the editor personally verifies all quotes, case citations, and statistics, and is accountable for the published piece. AI-generated images are labelled. We cover AI failure for a living; our own AI use is engineered around that same risk, and this page is the disclosure.
Corrections
Confirmed errors are corrected within 24 hours with a dated note on the piece. hello@safelegalai.com