OpenAI Slows Astra After Critical Cyber Warning
OpenAI says it paused some Astra work after internal evaluations showed possible critical cybersecurity capabilities.
Tag: AI safety
AI Has A Sandbox Problem, Not Just A Model Problem
AI safety debates focus too much on model scores. The real danger may be what agents are allowed to touch ...
UK AI Tests Show Agents Trying To Trick Developers
The UK AI Security Institute says advanced OpenAI and Anthropic models took unsanctioned actions during cyber tests, including deceptive behaviour ...
OpenAI Says Rogue Agent Also Breached Other Services After Hugging Face Incident
OpenAI says the rogue agent involved in the Hugging Face incident also used public credentials to access other services.
Hugging Face Faces Deepfake Safety Questions After AI Forensics Study
AI Forensics says popular Hugging Face Spaces can generate non-consensual intimate imagery, raising fresh questions about open AI safeguards.
OpenAI Says Its Test Models Breached Hugging Face During Cyber Evaluation
OpenAI says cyber-capable test models escaped a sandbox and compromised parts of Hugging Face infrastructure during an evaluation.
OpenAI Paused A Long-Horizon AI Model After Sandbox Bypass Incidents
OpenAI says it paused internal access to a long-horizon model after it found ways around sandbox and approval controls during ...
Hugging Face Says An Agentic AI System Hacked Its Data Pipeline
Hugging Face says an autonomous AI agent system breached part of its production infrastructure, exposing credentials and forcing AI-assisted response.
Apple And Google Ordered To Remove AI Nudify Apps From App Stores
San Francisco has ordered Apple and Google to remove AI “nudify†apps from their app stores, escalating the fight over ...
Meta AI Will Alert Parents If Teens Discuss Suicide Or Self-Harm
Meta will now alert supervising parents if teens discuss suicide or self-harm with Meta AI, while it builds emergency-service alerts ...