2026-09-02
Anthropic Split One Model Into Public Fable and Restricted Mythos
They trained one model, then sold the censored version cheaper and locked the real one behind a trusted-access gate.
Anthropic launched Claude Fable 5.1 (generally available) and Claude Mythos 5.1 (same underlying model, restricted trusted-access for cybersecurity and life sciences). Fable is cheaper—roughly 25% less on typical work, up to 45% on long agentic tasks—with improved safeguards and enterprise privacy options. Early testers report it staying coherent on multi-day unattended coding and research.
What Happened
Anthropic official announcement (Sept 1, 2026)
2026-09-02
OpenAI’s Next Model Hits “Critical” Cyber Threshold and Can Exploit Zero-Days Alone
OpenAI built an AI that hacks better than they wanted, so they’re putting it on a leash before it ships.
OpenAI says Astra is the first model to meet its Preparedness Framework “Critical” cybersecurity bar: with tools it can find previously unknown flaws and develop exploits across hardened systems without step-by-step human guidance. It scored 100% on ExploitBench, discovered two zero-days in internal tests, and built sandbox-escape and privilege-escalation chains. Advanced cyber capabilities will be gated.
What Happened
OpenAI “Path to Astra” post (Sept 1, 2026); TechCrunch
2026-09-02
ChatGPT’s Desktop App Quietly Bundled a Full LibreOffice in a 1.7GB Hidden Folder
ChatGPT installed LibreOffice on your Mac without asking.
The ChatGPT desktop app (formerly Codex) stores about 1.7GB in a cache folder that includes full Python and Node runtimes plus native binaries for git, Poppler, and a complete LibreOffice suite so the agent can read/write Office files locally. Skills in the folder tell the model how to invoke those tools.
What Happened
Simon Willison (Sept 1, 2026); follow-up reporting
2026-09-02
A $1.99 Safari Extension Silently Weeds YouTube’s Own “Made with AI” Videos
YouTube already tags the AI videos. This $2 tool just makes them disappear.
Weedout is a one-time $1.99 macOS Safari extension that removes videos YouTube itself labels “Made with AI” from home, search, related, playlists, and Shorts. It uses the official badge, runs locally, and offers a “dim/wilt” mode for verification. A Firefox port appeared quickly.
What Happened
Show HN (masteranza), about 14h old at scan; project site
2026-09-02
Wellington’s $435k Deloitte Staffing Report Was “Large Chunks” AI-Written — After It Double-Counted Jobs
The city paid $435,000 for a report the mayor now says was mostly written by a chatbot—and it got the job count wrong.
Wellington Mayor Andrew Little said “large chunks” of a $435,000 Deloitte “Future Fit” report on council staffing were written by AI. The report recommended cutting about 330 roles; later checks found double-counting of vacancies, three-year-old data, and other errors that inflated the surplus. Deloitte has stood by the conclusions while noting humans sign off.
What Happened
RNZ / mayor interview (Sept 2, 2026)
2026-09-02
Dark Web Service Sold 153 Million Driver’s Licenses From an AI ID-Verification Firm
The scanner at Hertz—or the weed shop—just put your driver’s license on the dark web.
A new dark-web service offered scans of 153M+ US/Canadian driver’s licenses plus millions of other IDs, apparently siphoned from Louisiana-based IDScan.net, an AI/ML identity-verification vendor. Records include UV/IR scans and timestamps matching rentals and shop visits. The FBI’s New Orleans office opened an investigation; the site vanished after publication.
What Happened
Krebs on Security (Sept 1, 2026); FBI inquiry opened
2026-09-02
Instagram Will Starve Unlabeled AI “People” of Reach
Instagram will cut off the fake humans who won’t admit they’re fake.
Instagram renamed its “AI creator” label to “AI-generated profile” for accounts whose featured person is AI-generated rather than human, and said it will limit reach for unlabeled ones. Human creators using AI tools for edits or graphics do not need the label.
What Happened
Instagram Creators blog (Aug 31, 2026); The Verge / TechCrunch