AI model auditing
AssessTechniques
Methods for inspecting and testing AI systems to understand behavior, capabilities, and risks.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by research-stage coverage, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 8Hacker News·10/13/2026researchStudy Claims AI Models Can Hide Capabilities
A preprint on arXiv argues that AI models can provably conceal arbitrary capabilities under certain conditions. The result suggests that standard evaluations may fail to reveal what a model is actually capable of doing, raising concerns for auditing and safety testing.