This week in legal AI safety — Issue #0 (sample)

This is a format sample — Issue #1 ships at launch. Each item below is real and primary-sourced; the selection illustrates a typical week.

Incidents

England & Wales — the Divisional Court’s warning stands. The referrals in Ayinde v Haringey remain the reference point for UK practitioners: barrister to the BSB, solicitors to the SRA, and an explicit warning that the court’s restraint on contempt should not be read as precedent. We track the disciplinary outcomes as they land.

Canada — costs against the lawyer personally. Zhang v Chen set the Canadian template: two ChatGPT-invented cases, costs ordered against counsel personally, other files reviewed, LSBC investigation. Canada is now the second-busiest jurisdiction in the global incident record.

Regulator watch

Australia operates the strictest evidence rule in the common-law world — NSW’s SC Gen 23 prohibits generative AI in affidavits outright and requires a declaration that it was not used. Our explainer maps it against the disclosure-based regimes in Canada and the Federal Court of Australia.

One number

3 jurisdictions, 3 regulatory referral systems, 1 pattern. Every tracked incident this issue involved the same failure: citations filed without verification against a primary source. The tool changes; the duty doesn’t.

---

SafeLegalAI is published by the team behind LegalAI Space. LegalAI Space is never scored on this site — ownership and standards.