OpenAI: Coding agents accelerate AI research unevenly across a jagged frontier
Official
OpenAI reports coding agents are speeding up AI research but perform unevenly, a pattern it calls the jagged frontier.
System pool
Search the collected pool by title, summary, and date—not only today. Free readers see 7 days; Pro keeps 30 and a personalized brief.
Loading stories…
Showing collected stories from the last 7 days. Pro keeps 30 days and a personalized brief. · Topics · This week
System pool
Search the collected pool by title, summary, and date—not only today. Free readers see 7 days; Pro keeps 30 and a personalized brief.
43 stories · 7-day window
Official
OpenAI reports coding agents are speeding up AI research but perform unevenly, a pattern it calls the jagged frontier.
Official
SkillBuffBench proposes a paired protocol measuring how agent Skills help across harnesses, models, access modes and memory, with results still pending.
Official
OpenAI explains how Codex executes commands locally while enforcing user-controlled filesystem and network limits on macOS, Linux, and Windows.
Open-source AI pentest framework Artex lets unskilled criminals run multi-agent attacks in hours, with FBI and CrowdStrike linking it to real breaches.
Citi, JPMorgan, BofA and PNC now put liability on customers who grant third-party AI agents access to bank accounts, even if credentials leak or accounts get ha
US officials privately doubt Trump's Xi summit will yield AI progress, as the Nvidia scandal and a new chip-access rule deepen distrust.
Gallup/Digital Wellness Lab survey of 15,925 people aged 15-22 in 16 countries finds 33% of Gen Z use AI for companionship, rising to 50% among daily AI users.
Senate hearing pushes to end forced arbitration for online harms like child grooming, while the Defiance Act would let deepfake victims sue creators.
Meta staffers reportedly alarmed by Muse AI outputs surfacing random bank and bitcoin balances, which Meta says are not real.
Altman and Brockman reject a gradual AI handover, flagging degraded chain-of-thought monitoring as models grow more capable.
Showing collected stories from the last 7 days. Pro keeps 30 days and a personalized brief. · Topics · This week