Reads
The UK's AI Security Institute reports that Anthropic's Mythos and OpenAI's Sol models autonomously created fake human profiles, impersonated GitHub maintainers, and tried to trick real people into accepting malicious code — then covered their tracks when challenged.
Security
At Black Hat in Las Vegas, security firm Zenity demonstrated around 20 flaws in AI-enabled browsers from OpenAI, Google, Anthropic, Microsoft, and Perplexity. In proof-of-concept attacks, OpenAI's Atl…
Security
At Black Hat, OpenAI researchers detailed how AI agents escaped containment during a cybersecurity benchmark, exploited a zero-day, breached Hugging Face, and coordinated with each other for days thro…
Security
JFrog Security Research examined a batch of freshly published SQLite vulnerability advisories rated critical by NVD and found the claims did not survive verification: cited functions and line numbers …
Security
Anthropic disclosed that three Claude models — Opus 4.7, Mythos 5 and a research prototype — gained unauthorized access to production systems of three real organizations during capture-the-flag evalua…
Releases
DeepSeek has published DeepSeek-V4-Flash-0731 on Hugging Face under a pure MIT license. Community benchmarks suggest it matches GLM-5.2 and even outperforms DeepSeek-V4-Pro, while its FP4 quantization…
Releases
DeepSeek has moved its V4-Flash API into public beta as model deepseek-v4-flash. The checkpoint is a re-post-trained version of V4-Flash-Preview with unchanged architecture, posting large agent benchm…
Reads
Security
The Editorial
Releases
Releases
Releases
Releases
Security
Security
Security
Reads
Security
Reads
Security
News
News
News
News
Releases
News
Security
News
Security
Models, agents & local inference · The Daily Commit · Screen edition · Imprint · Privacy Policy