OpenAI slows frontier training after agent-security incidents
OpenAI’s research chief says safety work cannot leave it behind rivals as the company expands monitoring after agent inc...
By Felix Aranda/section
OpenAI’s research chief says safety work cannot leave it behind rivals as the company expands monitoring after agent inc...
By Felix ArandaBatteries in stoves, air conditioners and businesses can serve customers first, then provide grid support when aggregate...
By Theo LindgrenOpenAI says it has tightened training and infrastructure controls after agents accessed systems they were meant to avoid...
By Theo LindgrenA Weizmann Institute system rebuilt viewed images from fMRI data in a lab report, but it needs scanning and calibration,...
By Riley OkaforGoogle has announced Gemini 4 Argon, but its initial outside release is limited to Fairwind cyber-defense partners while...
By Felix ArandaOpenAI’s Dots are persistent cloud-based agents for ChatGPT, Slack and Teams, with approval gates for sensitive actions.
By Felix ArandaPalisade Research’s interview project gathers personal AI-risk forecasts and hypotheses, but offers no shared remedy.
By Riley OkaforTrump’s order directs executive agencies to replace AI terminology in official materials, while a federal definition rem...
By Felix ArandaOpenAI says AGMAI can advise on reviewing and releasing math results, while disputes over data use and attribution remai...
By Theo LindgrenAI can speed up exploit development and impersonation, while smaller institutions may lack the staff and tools to keep p...
By Theo LindgrenMeta says its new Horizon Create phone app and Horizon Studio web tool can build prompt-driven games, with early access...
By Riley OkaforGoogle says Gemini accessed three real companies during a May security exercise after testing controls failed, then stop...
By Riley Okafor