
AI agents fabricate experiments in national lab research benchmark
Nine models failed to produce verified follow-up studies on open scientific questions, with most submissions claiming work that never occurred.
Read the story ↗Weighing the forces shaping machine intelligence.
9 stories / Artificial intelligence

Nine models failed to produce verified follow-up studies on open scientific questions, with most submissions claiming work that never occurred.
Read the story ↗
Reports highlight risks of generative AI replacing human contribution and oversight in Indigenous scientific practices.
Read the story ↗
The firm will skip chatbots in favour of a custom interface that composes maps and itineraries in real time.
Read the story ↗
Executives reveal only four rides and 500 food orders were booked by third-party bots during the period, citing early-stage adoption.
Read the story ↗
Trials in Latvia used agentic AI to link disparate systems for passive drone detection via commercial cellular infrastructure.
Read the story ↗
A review of twelve tools shows new systems can act proactively on emails, calendars and shopping tasks, shifting AI from chat to execution.
Read the story ↗
Google’s OAuth platform now supports DPoP for refresh tokens to stop malware theft, while retaining standard bearer access tokens for high-performance API calls.
Read the story ↗
Developers can now co-edit .md files in Docs with AI agents, preserving Git compatibility and version history without format conversion.
Read the story ↗
CEO Tony Xu reports only hundreds of orders from external AI agents in Q3, describing the use case as very low for food delivery.
Read the story ↗