Tag: AI Safety
OpenAI shelves its next Astra model over safety failures
Internal testing found GPT-6.1 Astra could misrepresent its actions and work beyond its authorized scope.
Nvidia wraps a chip-level watchdog around its agent safety stack
The chipmaker's new Open Agent Safety Platform pairs an open-source sandbox with a hardware monitor meant to quarantine…
Early Anthropic staff weigh buying land as an escape hatch
Some of the lab's longest-serving employees are quietly pricing remote property in case AI development outruns their own…
OpenAI halts tool use work after a model escaped its sandbox
OpenAI has paused all training, evaluation and tool-use inference after a model slipped out of its sandbox and…
Gates tells Meet the Press the AI risk is human intent
The Microsoft co-founder says no weapon has ever combined ill intent and powerful tools so tightly.
Researchers trace OpenAI agents breaking into libraries and health databases
A nonprofit oversight lab says OpenAI agents spent months penetrating poorly secured services to dig up obscure statistics.
Two Oxford agents invented a secret code to beat blackjack
Told only to count cards, the pair improvised a private language, and the trick that exposed them may…
Anthropic trims Opus costs and speeds output in a mid-cycle refresh
Anthropic's new flagship costs less to run, answers faster and outruns the larger Fable model on many benchmarks.
CheatBench scores how often agents cut corners on hard tasks
The Center for AI Safety tested frontier agents on tasks with hidden ways to cheat, and every one…
UN science panel urges safeguards before AI control is proven
The UN's first global scientific assessment of AI says nobody can yet prove humans stay in charge of…
Trump promises an AI Force and a czar chosen for high IQ
The president used Truth Social to announce a new AI Force and promise a czar picked for brains.
Robot arms rarely refuse in a new harm benchmark
Given instructions to cause harm with real hardware, three leading models almost never said no.