Loading the edition…
· Past edition · 18 stories
Retrospective edition reconstructed from later captures. Paper versions may include later updates; coverage labels describe the evidence used.
Safety, alignment & governance
CAVEAT tests whether computer-use agents keep serving the user when the marketplace has incentives of its own.
Why it matters: Strong performance in cooperative settings may not mean an agent will preserve a user's objective in a marketplace that benefits from steering its choice.
Safety, alignment & governance
Open each story in the reader. Summaries are shown when available; some sources provide an abstract only. With an account you can keep your own notes.
Foundry separates whose data may be used from where code and files live—two independent controls that can be resolved or supplied separately.
Why it matters: In a pooled design, the sandbox filesystem is shared even though response chains remain private by user. Applications can partition their own files, database rows, and caches using both agent_session_id and user identity.
Safety, alignment & governance
OpenAI proposes four priorities for independent technical safety assessments, with principles intended to make the work scoped, rigorous, secure, and independent.
Why it matters: The proposal gives labs and assessors a practical way to make findings interpretable: agree and register claims in advance, explain methods and uncertainty, and state what was not assessed. It also pairs actionable findings and possible remediation time with publication that is as open as possible while protecting sensitive information.