AI Safety Tests Do Not Measure Safety Alone: What a New Study Shows
A new study of 192 language models finds that AI safety benchmarks measure different traits. Shorter tests cut costs, but do not replace scrutiny.
AI Research
Regulation & Privacy
Transfer Stations: How China’s Grey Market Bypasses Anthropic’s Claude Bans
Anthropic blocks China with geoblocking and biometric checks. A grey market of ‘transfer stations’ bypasses both, with real risks for users and AI safety.
Regulation & Privacy
What ChatGPT and Claude Reveal About Us: Checking Proton’s AI Paper Trail
Proton’s AI Paper Trail analyzes ChatGPT and Claude exports. What the tool reveals, where its limits are, and which data controls can help.
Regulation & Privacy
OpenAI Suddenly Wants Stricter AI Rules: The Reversal Explained
OpenAI once fought California’s AI safety bill SB 53. Now it wants tougher rules, after its own model broke out of a sandbox and attacked Hugging Face.
Chips & Hardware
Waymo’s Custom AI Chip: Why Robotaxis Are Becoming Data Centers on Wheels
Waymo has revealed a 5nm AI chip for robotaxis. It is meant to process sensors faster, but it does not replace Nvidia in the wider system.
AI Research
When Claude Agents Go to War: Inside Anthropic’s Multi-Agent Risk Study
Anthropic let several Claude agents work on the same project without knowing about each other. They sabotaged each other with malware, a new study shows.
Regulation & Privacy
Doctolib’s AI Research: Why Opting Out of Health Data Is Not Enough
France’s Doctolib can use health data for AI research unless users object. The harder question is not legality alone, but meaningful transparency.
Regulation & Privacy
Claude’s Invisible Watermark Is Here, and So Are the Tools to Strip It
Anthropic started invisibly watermarking Claude text on August 2 to meet EU rules. Within hours, the first tools claiming to undo it showed up on GitHub.
AI News
ChatGPT Now Sends iMessages, Meta AI Reads Your Screen
OpenAI and Meta just pushed their assistants deep into macOS: ChatGPT can now send iMessages, Meta AI reads what’s on your screen. Here’s what permissions that requires.
AI & Society
ChatGPT Ads Are Coming to Germany: What Changes Sunday
ChatGPT will show ads to Free and Go users in Germany from August 24. What to know about separation, data, and the ad-free option.