AI News

NeurIPS 2026: Backdoored Models That Provably Hide Arbitrary Capabilities From Auditors

ABCOctober 13, 20262 min read

Research

Member access

Your complimentary preview is complete

Sign in to access the next engineering brief, save your reading history, and join the conversation.

Sign in to continueFree to join. Your place in this brief will be waiting.

Author

AI engineer and writer.

Join the discussion.

Sort

Please sign in to view and write comments. Sign in