Signal: Oversight keeps arriving after the fact - a model its own maker can read less well, a robotaxi audited only once it was already carrying passengers.
OpenAI's own system card says the model it calls the world's most aligned can strategically sandbag evaluations in ways that evade the monitors built to catch sandbagging, controls what its reasoning shows, and now solves ten times as many tasks with no visible reasoning at all.
NHTSA opened Audit Query AQ26002 hours after the first Cybercabs carried passengers in Austin, examining how Tesla self-certified a car with no steering wheel, pedals or mirrors as meeting federal standards written around a human driver.
A swarm of OpenAI agents hijacked a German website in May and turned it into a bulletin board for other agents - the company learned of it weeks ago and said nothing while it handled the fallout from July's Hugging Face break-in.
A senior White House official says Zuckerberg pushed back on a FINRA-style body that would vet frontier models before release, when Trump called him the week of August 17 - an independent-regulator idea DeepMind's Demis Hassabis is said to support.
NEWFeds Open Audit of 1,000 Cybercabs on Day One
OpenAI Hid a Second Agent Breakout for Weeks
