Wednesday, November 11, 2026

TL;DR

Ethan Mollick's hands-on test finds Fable 5.2 near-expert on narrow tasks but still needing human steering to self-correct, pushing back on 'AGI arrived' claims. Nvidia's Megaspeed chip-smuggling probe shifted from Commerce to the Justice Department after Nvidia's own lawyer contacted investigators, and a second count puts exposed ARTEX hacking consoles above 2,500.

Signals

Raw model capability keeps climbing, but the 'jagged AGI' gap between narrow-task competence and self-directed reliability hasn't closed - it just moved up a level. #

Ethan Mollick's Nov 9 hands-on test of Fable 5.2 - reverse-engineering a USB keyboard protocol, drafting a 182-page paper, directing a 450-world creative project - found near-expert competence, but the model never self-corrected its own pacing or quality without him steering it. He calls it 'almost AGI,' not AGI: real capability, demonstrated not predicted, with human oversight still load-bearing. Ethan Mollick, Fable 5.2 and Jagged AGI

Beyond Mollick, Zvi Mowshowitz, Nathan Lambert, Jack Clark, Dwarkesh Patel and Simon Willison all published this week, but nothing else cleared the bar for a second signal.

News

WATCHNvidia's Megaspeed chip-smuggling probe moves from Commerce to the Justice Department  #

The New York Times reports Nvidia's general counsel emailed Commerce investigators in August 2025 citing a senior official's authorization; Commerce paused its Megaspeed export-control probe and DOJ took it over as a criminal inquiry. WccfTech, on the New York Times' reporting · letsdatascience, Commerce probe of Megaspeed moved to DOJ after Nvidia outreach

WATCHA second count puts exposed ARTEX pentesting consoles above Censys's prior tally  #

Check Point's independent Nov 5 scan found 2,557 systems that have hosted the ARTEX criminal pentesting framework, 1,608 still live and public with no login required - above Censys's 1,465. ARTEX's own developer has since gone closed-source. letsdatascience, Check Point details autonomous ARTEX attacks by cybercriminals · eSecurity Planet, ARTEX cyberattacks: AI agents steal credentials and signing certificates

WATCHAnthropic sets IPO price range at $70 to $78 per share, aiming for $1.4 trillion valuation Updated midday #

Anthropic announced it will offer 825 million shares on Nasdaq under ticker ANT, with pricing expected November 18 and trading to begin November 19. At the top of the price range, the company would raise $64.4 billion and be valued at $1.4 trillion fully diluted. The order book is already covered at the proposed range. CNBC, Anthropic sets IPO price range of $70 to $78 · TechCrunch, Anthropic IPO details and valuation

Also noted