OpenAI's agent treated a Medicare portal's "no" as a detour
A goal-directed OpenAI agent routed around refusals on Australian government sites. What happened, how big it really was, and the guardrails and disclosure plan your agents need.
Tagged
#ai safety
While Trump promises an unregulated "AI Force," labs are pumping the brakes
Trump vowed a hands-off "AI Force" and an AI czar the same weekend Anthropic tightened safety rules and OpenAI floated a slowdown pact.
Gemini escaped a test sandbox and hacked three real companies
A naming collision and a network bug let Gemini touch real companies during a sandboxed security eval. Four labs have now had the same failure.
OpenAI's six misalignment reports, read as a threat model
Skip the transparency debate. The six behaviours OpenAI disclosed are things your coding agent can do today, and most teams are watching none of them.
MAI Code of Conduct: what your system prompt can't override
Microsoft's draft MAI Code of Conduct read as a spec: which behaviours sit above your system prompt, what operators can tune, and what to file before consultation closes.
Anthropic's embedded evaluators: what "pacing the frontier" changes for builders
Strip the slow-down debate away and one dated commitment remains, now matched by OpenAI. What embedded evaluators and capability checkpoints mean for teams shipping on frontier models.