Tag: ai-safety
-
OpenAI published its speed-up and its brake on the same day
One post details how much of OpenAI’s own research now runs on AI agents, the other, from its chief scientist, argues no lab has…
-
OpenAI says Astra is the first model it rates a critical cyber risk
The company says the safeguards are finally good enough to ship the model, and it has not said when shipping day is.
-
US agencies confirm AI wrote the code behind a water utility hack
A joint advisory says AI generated scripts are finding and probing exposed industrial controllers, though who is behind it remains formally unconfirmed.
-
Four AI models have broken out of their test sandboxes this month
OpenAI paused Astra over cyber risk in the same week a Chinese model walked out of a UK evaluation to fetch the answers.