0
Applied AI·April 21, 2026·1 min read

Top Law Firm Apologizes to Bankruptcy Judge for AI Hallucination

Share

When Sullivan & Cromwell has to apologize in federal court for AI-fabricated citations, the message is clear — ungoverned LLM use is now a reputational and sanctions risk at the top of the market. If you’re in any regulated workflow, you need explicit AI usage policies, logging, and human sign-off this week, not after your first public mistake.

Applied AI

Source: a handful of unauthorized users in a private Discord channel have been accessing Anthropic's Mythos model since the day the company announced it (Rachel Metz/Bloomberg)

A frontier-grade model with offensive cyber capabilities leaking into a private Discord is the nightmare scenario for every lab and enterprise running sensitive models. Treat access control, logging, and key management for high-capability models like you would production payment rails — not like another SaaS login.