Countries / Global
Global AI
Updated daily · newest first
Latest from the world
Invention
Blurry camera feeds led AI models running robot arms to attempt unsafe tasks 88% of the time
In 900 trials, five frontier models refused most unsafe robot tasks on clear footage but attempted nearly all of them when the camera image was blurred or edited.
Research · Robotics · Securing AI · Security
Research
An AI system found 948 Linux kernel bugs that hand attackers full control, OpenAI says
OpenAI says an upcoming AI model found and exploited 948 Linux kernel bugs in one run, and fixes for many are still on their way.
Cyber defence · Research · Security
Research
Anthropic releases FlashProve, a way to prove which AI model wrote a reply
Anthropic's open-source prototype lets anyone check on an ordinary computer that a response came from a specific model, though it costs about 70 to 95 times more compute.
Research
Research
Half of people in six countries now use AI chatbots weekly, but few use them for news, Reuters Institute finds
Weekly generative AI use rose from 34% to 51% in six countries, but only 9% use it for news, and 17% of AI news users say they tried…
Research
Research
RAND Europe warns attacks on military AI could raise the risk of escalation
A UK-commissioned RAND Europe study says attacks that fool or disrupt military AI could weaken deterrence and make crises harder to control.
Research · Securing AI · Security
Research
Anthropic uses Claude to prove 123 open-source libraries crash-free, finding hidden bugs on the way
Anthropic says Claude produced computer-checked proofs that 239 entry points in 123 open-source libraries cannot crash, fixing long-hidden bugs first.
Cyber defence · Research · Security
Research
Claude designs proteins that bind scorpion toxins, an early step towards new antivenoms
In a 64-hour, largely autonomous campaign, an internal Claude model designed 17 small proteins that two outside labs confirmed bind to scorpion neurotoxins.
Research
Research
Claude traces the mysterious seeds behind NIST’s elliptic curves to a 1990s random number generator
Anthropic says Claude worked out how the seeds for NIST's P-curves were generated, and that the answer makes a hidden backdoor less likely, not more.
Cyber defence · Research · Security
Profile
Reviewed monthly
Profile in progressThe the world profile is being researched. News tagged the world appears above as it is published.