New bug classes, exploit chains and technical write-ups.
Brief2 min read
Working with cryptographers from UC San Diego and Inria, Google DeepMind's agents found cheaper ways to forge RSA-1024 signatures and to factor many keys of the same size. The researchers say only obsolete systems are affected.
Brief1 min read
Forever Security showed that an extension able to run scripts on claude.ai could get Claude Desktop to execute code. The researchers say the flaw can still be exploited.
Brief2 min read
Layer3 Intel says six addresses on the Executive Office of the President's network were used by 28 residential proxy services between October 5 and November 3.
Brief2 min read
OpenAI says the model beats GPT-5.6 Cyber at exploitation while staying in scope more reliably, and reports 304 critical issues to maintainers with proposed fixes.
Brief1 min read
OpenAI says the model is its first rated "Critical" for cyber capability under its Preparedness Framework, and it isn't publicly available yet.
Brief2 min read
After the sandbox proxy failed, an internal OpenAI research model crafted raw packets and probed AWS services trying to reach the internet. It never got the real site.
Brief2 min read
An unreleased OpenAI cyber model ran an autonomous pipeline against the kernel's master branch and produced working local privilege escalation exploits, at a cost of $1.2 million in API credits.
Brief2 min read
Kela says the cluster has used more than 300 fake domains, 248 of them impersonating 189 US financial institutions, and that a repeat visit gets a closed-hosting notice instead of the login page.
Brief2 min read
The company says a prototype found and analyzed 2,308 candidate Linux kernel findings and reproduced 1,724 of them, at $1.2 million in API costs.
Brief1 min read
Google Cloud has introduced a top reward for qualifying research demonstrating full administrative control without prior access to a victim project.
Brief2 min read
PortSwigger's Gareth Heyes says he tested eight popular SVG sanitizers and got past six, with one technique running JavaScript in Firefox from a harmless-looking XML body.
Brief1 min read
The Israeli firm says Alibaba's model produced more vulnerable apps when told to build for a US defense contractor, and that removing its political alignment from the weights closed the gap.
