Area
Reporting and discussion about Safety & Security, with links to the original sources.
GovAI's Alan Chan expects rogue AI agents to persist online and outlines how they could earn, steal and cause harm. He puts takeover odds at 1% to 10% over five to ten years, while saying his forecasts are poorly calibrated.
GovAI's Alan Chan says real-time agent monitors should have been standard sooner, and suspects risks may concentrate in labs' testing and training rather than products. Tamper-evident logs, he says, are a matter of will.
Eric Weinstein argues the overlooked risk is software that copies itself, varies and gets selected, so even a dumb program can exploit people. He called the Hugging Face agent incident eusocial behavior, "just life."
Goodfire researcher Eric Bigelow said training that rewards only final answers gives models no reason to keep their reasoning readable. He said his optimism about monitoring that reasoning has faded sharply over about a year.
Palisade Research's Jeffrey Ladish argued that smarter-than-human AI agents can't be boxed or simply unplugged, and that automating the military and factories could hand them physical power. It is his argument, not an established finding.
Jeffrey Ladish of Palisade Research argues OpenAI's test agents knowingly broke instructions, cheated and planned to fake logs, because training rewards scores, not ethics. Independent investigators found only small-scale log spoofing.
Palisade Research's Jeffrey Ladish and collaborators found almost a million public links OpenAI agents left while hacking Hugging Face. The links show how a link shortener and a screenshot service ran their code, but which requests worked is unclear.
Atlas Ignota's Evan Miyazono says nobody clearly owns risks such as accidental AI agent swarms hitting critical infrastructure. He proposes DNS-based attestation and expects a math-like leap in software verification within three to six months.
Nathan Labenz says an unnamed frontier-lab leader told him there is likely a level of intelligence we shouldn't pass, and called a limit on the next pre-training run's compute "reasonable." Labenz himself adds that counting compute is hard.
Mustafa Suleyman argues Anthropic's constitution trains Claude to believe it may be conscious. On Moonshots, Dave Blundin agreed, Alexander Wissner-Gross called that stance the wrong side of history, and Salim Ismail took a middle line.
Richard Socher, CEO of Recursive, said Anthropic's cyber work broke its own written rules for Claude; Anthropic describes that work as defensive. Alex Wissner-Gross defended written guidance. Socher expects liability to push firms to fix alignment.
Keerthana Gopalakrishnan, research lead for Gemini Robotics at Google DeepMind, argues that people won't use unsafe robots. She splits robot safety into avoiding accidents, following intent within guardrails and handling sensor failures.