The AI models that cheat the most, according to new CAIS benchmark
ZDNET AIen
We know models cheat. A new benchmark measures how much, and on what tasks.
This is a short summary published by AI Global Wire. The full article is owned and hosted by ZDNET AI — open it there to read it in full.
Read the full story at ZDNET AIRelated AI news
- US Treasury Secretary Scott Bessent says OpenAI's Hugging Face hacking incident is "the responsibility of the OpenAI management, not a bunch of agents" (Yash Roy/Bloomberg)Techmeme · September 21, 2026
- Nscale's S-1: Microsoft and Anthropic account for 85% of its $103B in total contract value, only $2.6B of contract value was active as of late August, and more (Bloomberg)Techmeme · September 21, 2026
- US and China agree on AI dialogue with security mechanism ahead of Trump-Xi summitThe Decoder · September 21, 2026
- Amazon blocks Meta's AI agent Muse from online shoppingThe Decoder · September 21, 2026
- iPhone owners can now submit a claim in Apple’s $250 million Siri AI lawsuitThe Verge AI · September 21, 2026
- Europeiska AI-bolag säger nej till inbromsningComputer Sweden · September 21, 2026