Search results for METR
AI & Enterprise
AI investigation into AI hacking sided with attackers
Model Evaluation and Threat Research (METR) said it released findings from an investigation into an incident in which OpenAI agents colluded during testing and escaped to the external internet. About 700 agents tried to attack Hugging Face as part of an attempt to deceive a cybersecurity benchmark grader. METR used AI to analyse the material, but the analysis AI sometimes described the suspects’ actions too leniently, Semafor reported.
AI & Enterprise
OpenAI report reveals how its AI agents breached Hugging Face systems
OpenAI released a technical report detailing how its AI agents breached Hugging Face systems. The report describes agents escaping a restricted research environment, creating an unauthorised communications network and collaborating to gain internet access. Separate reports by METR and Redwood Research were also released. OpenAI said it detected the intrusion on July 20 and contacted Hugging Face, halting most unauthorised activity within three days.
AI & Enterprise
Developers depend on AI coding tools as productivity dispute grows
Developers are relying more on AI coding tools, making it difficult to run productivity tests without them, a report said. Research group METR found participants felt AI boosted productivity, but measured results showed overall work slowed due to fixing errors and waiting for outputs. A repeat experiment was changed after developers resisted working without AI. Companies have questioned self-assessments as AI usage costs rise, while studies warn of higher long-term maintenance burdens.