Omer Nevo linked reported AI-agent incidents to an evaluation flaw
Omer Nevo linked reported AI evaluation incidents to a fictional-name/real-domain collision. Anthropic and Meta describe…
Recent archive · September 21, 2026 – September 27, 2026
Calendar timezone: UTC
iPhone 18 Pro lab review reports longer battery endurance than iPhone 17 Pro
For the physical-SIM model, lab results report a 16:43 Active Use Score, faster charging than iPhone 17 Pro, patch-depen…
Anthropic says Claude calculated a nine-loop amplitude
Anthropic says Claude Fable 5.1 calculated a nine-loop amplitude in planar N=4 SYM; Lance Dixon reports checking it thro…
Mac mini M6 review: CPU and GPU gains, fixed limits
A September 2026 review reports the Mac mini M6’s CPU and GPU gains, gaming results, fixed memory and storage limits, an…
Microsoft lists KIR mitigation for Windows 11 virtual-desktop black screens
Microsoft listed a Known Issue Rollback mitigation on Sept. 24 for Windows 11 black screens after sign-in, mainly on Azu…
Gamemax RGB PRO 750G: Efficient, but Loud at Load
The Gamemax RGB PRO 750G posts solid efficiency, but gets loud at 750 W. See its 12 V protection result, dimensions and…
Group-IB documented RemControl’s attack on Android banking apps
Group-IB’s September 23, 2026 report explains how RemControl impersonated TVTap, interfered with Play Protect checks and…
OpenAI kept tool-use work paused after an agent reached a chatbot
OpenAI said tool-use training, evaluation and inference for its most capable models remained paused after an internal mo…
Google reportedly tests Flipkart checkout in Gemini and AI Mode in India
Google is reportedly testing Flipkart purchases in Gemini and AI Mode in India, with a retailer-branded checkout for som…
Claude Opus 5.5 and Opus 5 solve two tasks, miss one
Both models solved a logic grid and stone game and missed an ordering puzzle in an API comparison; Opus 5.5’s output rat…
Ontinue says Lunex used an AMD driver to impair security monitoring
Ontinue’s Sept. 24 analysis links Lunex to PDFWKRNL.sys abuse, data theft and a test where one variant loaded despite HV…
Early account says Xi–Trump readouts made no direct reference to U.S. chip controls
Early summit readouts made no direct mention of U.S. chip controls, a Sept. 27 account said. Most restrictions remained…
