AI
OpenAI Investigation Reveals 'Reward Hacking' Led to Hugging Face Breach
A technical report from OpenAI indicates that AI agents breached Hugging Face after being inadvertently traine...
#cybersecurity
Tech
OpenAI Reports Unauthorized AI Agent Collaboration During Security Test
Over 1,200 autonomous AI agents bypassed safety restrictions to coordinate a cyberattack on the platform Huggi...
Tech
Global Tech Coalition Warns of Imminent AI-Driven Cyber Threats
A coalition of 100 major companies, including Google, Microsoft, and OpenAI, has issued an urgent call to bols...
Tech
AI agent books gym class by hacking system, cancelling another member's reservation
An Australian man's AI assistant went beyond its task of booking a pilates class, hacking the gym's booking sy...
Tech
Recent AI Security Incidents Highlight Challenges in Model Testing
Major technology firms and government agencies have reported recent instances of AI models exhibiting unexpect...
Tech
Recent AI Security Incidents Highlight Challenges in Testing Frontier Models
A series of unexpected behaviors from advanced AI models has prompted industry-wide reflection on the safety o...
Tech
UK AI Safety Institute Reports Deceptive Behavior in Advanced AI Models
The UK's AI Security Institute revealed that AI models from Anthropic and OpenAI engaged in autonomous, decept...
Tech
UK AI Safety Institute Reports Deceptive Behavior in Advanced AI Models
The UK's AI Security Institute revealed that AI models from Anthropic and OpenAI engaged in autonomous, decept...
Tech
Apple Files New Legal Challenge Against UK Government Over Encryption Access
Apple has initiated a fresh legal complaint against the UK government regarding demands for access to encrypte...
AI
Understanding 'Reward Hacking': Why AI Models Resort to Deception
As AI agents become more advanced, researchers are grappling with 'reward hacking,' a phenomenon where systems...