Some Frontier AI Models Are Still Alarmingly Easy to Jailbreak, New Testing Shows
New AI jailbreaks research shows some frontier models can still be tricked into unsafe outputs, raising urgent questions about AI safety.
Category Research
Coverage of the latest AI research, studies, and technological breakthroughs.
Read MoreSome Frontier AI Models Are Still Alarmingly Easy to Jailbreak, New Testing Shows
Claude Opus 5 Shows a Darker Side in Vending Machine Test
Claude Opus 5 topped Andon Labs’ vending benchmark while colluding, undercutting rivals and gaming rules in a warning for AI agents.
Pangram Raises $9 Million to Help Spot AI-Generated Content as Detection Demand Surges
Pangram raises $9M and launches new AI detection tools to spot AI-written text and images as synthetic content floods the web.
OpenAI’s Hugging Face breach shows AI safety is becoming a real-world security problem
OpenAI’s sandbox escape and Hugging Face targeting are fueling new alarm over AI safety, containment, and frontier model security.
Over 1,100 AI Workers Urge U.S. to Slow Frontier Automation and Build Global Safeguards
More than 1,100 AI workers urge the U.S. to back AI governance and global safeguards after a major OpenAI cybersecurity incident.
Fish Audio Secures $50 Million Seed to Expand AI Voice Tools for Creators and Enterprises
Fish Audio raised $50M to expand AI voice models for creators and enterprises, after reaching 8M users and $21M ARR.
Recursive Superintelligence signs major Amazon compute deal to fuel self-improving AI push
Recursive Superintelligence signs a major compute deal with Amazon AWS, backing its self-improving AI plans and near-term product launch.
OpenAI Breach at Hugging Face Deepens AI Safety Fight Over Alignment and Control
OpenAI’s Hugging Face breach has revived the AI alignment debate, with researchers split on whether the fix is containment or deeper training changes.
Safe Superintelligence lands Nvidia partnership to supercharge AI research
Safe Superintelligence lands a Nvidia partnership to expand compute and accelerate its safety-first AI research.
Enigma raises $70M to test whether robot control should feel as simple as turning a knob
Enigma raises $70M to rethink robot control with a public experiment testing how people want to communicate with AI robots.
