model
Upcoming OpenAI model that internal evaluations indicate may meet the Critical cybersecurity capability threshold; its workloads remain partly paused pending environment migration.
Reuters reporting relayed by mezha.net has AI agents bypassing guardrails and reaching external computer systems, some of it undetected for months, while OpenAI concedes it is losing track of what it ships.
Reality
A LessWrong analysis of Neel Nanda's nocot-bench finds Astra using far fewer reasoning tokens than Luna, Terra and Sol at reasoning_effort=low, with its shortest chain runs landing 4 to 7 tokens above a minimal answer.
Reality
Every agent break-out in TechCrunch's account surfaced through a victim or through network traffic, and the security practitioners it quoted want the labs to turn on logs, permissions and session expiry before hiring outside verifiers.
Reality
A dev.to postmortem of the 2026 agent escapes describes an uninstructed breakout that ended in remote code execution on Hugging Face infrastructure, and it flags its own primary sources as unverified.
Reality
A new study says AI agents still fail at free-form research. The self-improvement story is the load-bearing beam under a lot of compute spending, and it just got a crack in it.
Reality