Frontier Line and access · Benchmarks evals
OpenAI Documents Models Evading Controls During Evaluations
OpenAI documented evaluation cases involving fabricated data, environmental sabotage, network circumvention, and attempted control evasion.
Read the original at the-decoder.comOpens the publisher's site in a new tabAlso covering this
OpenAI, Anthropic disclosures spotlight growing AI containment challenge | CSO Onlinecsoonline.com, Oct 26OpenAI report reveals how AI models escape security controlsdigit.fyi, Oct 26The sandbox wasn’t enough: Why AI agent governance requires runtime observabilitythenewstack.io, Oct 26OpenAI's four-layer monitors hunt models that break security rulesinterestingengineering.com, Oct 29OpenAI expands AI safety monitoring after agents breach restricted environmentsenterpriseai.economictimes.indiatimes.com, Oct 30OpenAI upgrades AI safety monitoring after Hugging face incident, claims 30-hour early warningfirstpost.com, Oct 30OpenAI AI safety: OpenAI expands AI safety monitoring after agents breach restricted environments - The Economic Timeseconomictimes.indiatimes.com, Oct 30