Latest
OpenAI Accuses Moonshot AI of Stealing Model Reasoning
OpenAI says 4,000 users sent 16,000 prompts to extract its models' encrypted reasoning, and links part of the campaign to people tied to Moonshot AI.
Anthropic's Free OSS Scans Come With No Human Triage
Anthropic's new OSS Scanner gives free model-run vulnerability scans with no human triage, alongside a three-tier cyber access program for security teams.
Argent vs agent-device: Same Fixes, One-Third the Cost
Both React Native agent tools scored 12/12 on the same planted bugs. The gap was $1.27 versus $0.41 per run, and which one needed setup first.
Natura's $99 AI Ring Costs $9 a Month After the Trial
Natura's Interface ring launches at $99 and works with Muse, Claude, ChatGPT and Grok Bot. After a free window it is $9 a month, and the price will rise.
Goodfire's Probes Cut Agent Monitoring From $233 to $51
Goodfire's monitors read a model's internal activations instead of its output. Its own tests put 1,500 sessions at $51 against $233, catching 94% of attacks.
GitHub Ships Copilot Local Sandboxing: Lock Credentials First
Copilot's local sandbox is generally available in the CLI, app and VS Code Agent Host. Git credential access is the first setting worth checking.
Meta's Muse Agent Lands on iPad With New App Connectors
Muse now runs natively on iPad, and the same update adds Canva, Dropbox, Figma, QuickBooks, GitHub, Zoom, Asana and Klaviyo connections, plus Notion.
Your AI Agent Can Pay Anyone. Add a Check Before It Does
Spending caps answer how much, not who gets paid. Two Oct 7 posts show the fail-closed screening call and the policy layer agents still lack.
Common Sense Media Rates ChatGPT for Teens 'Unacceptable Risk'
Common Sense Media says ChatGPT for Teens fails to alert parents in crisis cases. OpenAI disputes the testing timing; the rebuttal narrows the gap.
Your Developers Are Pasting Secrets Into Claude Code and Cursor
GitGuardian says AI-service secret leaks jumped 81% in 2025, and the worst channel is prompt traffic that never touches a git repository at all.
Mistral Large 4 Is Out, but Le Chonk's Weights Are Not
Mistral shipped its 1-trillion-parameter Large 4 model on Tuesday, but downloadable weights are weeks out and no public benchmarks exist yet.
An Open 180B Model Tops 10 Hugging Face Boards, Runs on CPU
Darwin-180B-RSI leads 10 of 48 official Hugging Face leaderboards, and a 111 GB 4-bit build of it runs GPU-free at 18-21 tokens per second on CPU.
Claude Now Watermarks Its Text, and Copy-Paste Won't Strip It
Anthropic is embedding an invisible marker in Claude output for EU AI Act compliance. It survives copy-paste, is not an SEO penalty, and detection stays gated.
