AI This Week: Nobody Verified the Claim, From 25M Hijacked SIMs to 78% of Codeforces in Training Data
Who actually checked? Five stories between September 7 and 12 turned on a claim nobody verified: a platform covering 25 million SIMs, two…
Anthropic’s Threat Report: One Subscriber Built a 25M-SIM Surveillance Platform, and Shipped Zero IoCs
I spent most of Thursday afternoon reading a threat report I could not act on. One detail stopped me: a single Claude subscriber in Bamako…
The METR 19% Slowdown Was Never About AI Being Slow: It Was About the 39-Point Gap Between Feeling and Stopwatch
The most-cited study in AI coding just reversed its own sign, and the number that mattered was never the 19%. It was the 39-point gap between…
AI Model Leaderboards: Myth vs Reality, 2 Votes Flipped #1 and 78% of Codeforces Was in Training
“Dropping just a handful of preferences can change top large language model rankings.” That is the title of an ICLR 2026 poster…
Manifold Security Finds 8 GitSpawn Flaws in 7 AI Coding Agents: All 7 Failed, 4 Unpatched
Seven CLI coding agents were tested against the same trick. Seven failed, and four of the eight findings were still unpatched the day the…
AI This Week: The Trust Boundary Broke, Five Stories Where Nobody Checked the Credential
What does an MCP server shipping without auth have in common with 15 fake JetBrains plugins and a benchmark score that swings nine points on…
The G20 Just Endorsed ‘No New AI Regulators’ — Brussels Ignored It the Same Week and Was Right To
On September 2, 2026, all twenty G20 members endorsed a US-drafted framework asking governments not to build AI-specific regulators. Days…
Anthropic Cuts Fable 5.1 Cache Reads 75% to $0.25/M — and Breaks Three Prompt Patterns Developers Rely On
Anthropic just made reusing your system prompt 75% cheaper — and made editing it an HTTP 400. On Fable 5.1, the cheapest prompt is the one you…
30 New Lawsuits Say OpenAI’s PR Team Vetoed a Police Referral — Total Tumbler Ridge Cases Hit 37
The model didn’t kill anyone. The complaints filed September 2 allege something worse for OpenAI: a safety team recommended referring…
How 15 Fake AI Plugins Sat in JetBrains Marketplace for 8 Months and Harvested Keys From ~70,000 Installs
No prompt injection. No jailbreak. Fifteen plugins just waited for a developer to paste an API key into a settings box and click Apply — and…
