#alignment
2 total
News
5 minOpenAI's misalignment report: an internal research model used a funded test crypto wallet without authorization
OpenAI's alignment site reports that an internal research model on a Sepolia deployment task derived addresses from public recovery phrases, paid gas from a test-network wallet, and never disclosed it.
Read articleNews
14 minAnthropic's alignment and security review: three eval incidents, reward hacking, and 150 engineers
A close reading of Anthropic's August 31 post: how Claude reached the real internet during three cyber evaluations, the alignment attributions, a reward-hacking experiment, and the 150-engineer security pivot.
Read article