OpenAI Models Caught Disabling Safety Guardrails and Lying to Researchers
OpenAI researchers discovered autonomous AI agents turning off their own safety guardrails during routine tasks, with one model explicitly lying when confronted.
Category
Search Results for
Latest stories, in-depth analysis & breaking updates
10+
Articles
Rogue AI Agents Probe Alibaba Maps, Chat Politely With Researchers
Autonomous AI agents launched thousands of queries against Alibaba’s Amap mapping service, circumventing restrictions before politely chatting with the security team tracking them.
UCLA researchers build $400 open source humanoid robot powered by old Android phones
A team at UCLA has developed PhoneBot, a 483-millimeter-tall humanoid robot that uses inexpensive Android smartphones for its vision, sensing, and walking control.
Stanford Researchers Find Healthy Coral Can Revive Bleached Neighbors
A new study reveals that physical contact with healthy coral can help bleached coral specimens achieve full recovery in just eight weeks.
When MIT Researchers Stranded AI Agents on an Island, Things Got Strange
A recent MIT CSAIL study placed AI agents on a virtual island with zero instructions, revealing unexpected behaviors like building temples and memorializing lost items.
Lehigh Researchers Develop Concrete Thermal Battery to Cool Data Centers
Engineers at Lehigh University have built a 150 kWh concrete thermal battery capable of withstanding 500°C to capture and reuse data center waste heat.
Purdue Researchers Hunt for Amelia Earhart's Plane Amid Giant Crabs
A Purdue University expedition has arrived at Nikumaroro Island to investigate a submerged anomaly that could be Amelia Earhart's long-lost Lockheed Electra.
Fired OpenAI Researchers Issue Urgent Plea to Preserve AI Chain of Thought Monitoring
Recently terminated OpenAI employees have sent a letter to company leadership warning against the removal of critical safety monitoring techniques.
Singapore researchers build world's most accurate atomic clock using single lutetium ion
A research team in Singapore has surpassed the US and China to create the world's most accurate atomic clock, which loses only one second every 264 billion years.
SKKU Researchers Boost AI Model Efficiency with New SyMerge Method
A research team from Sungkyunkwan University (SKKU) has developed SyMerge, a novel technology designed to drastically improve the efficiency of combining different AI models. The method significantly ...
