Black Friday comes early: nvestors skeptical Anthropic's pre-Thanksgiving IPO can hit a $1.5 trillion valuation after the AI selloff; one says closer to $1 trillion - The Wall Street Journal

AI and the 12x engineer: METR says Anthropic has three unreleased models, staff report roughly 12x productivity gains, and AI-assisted R&D compressed two years of progress into one - Business Insider

METR finds Anthropic likely improving model capabilities about twice as fast as its March 2024 to October 2025 pace - METR

Anthropic partners with Plaid to let US consumers connect financial accounts to Claude's personal finance product starting Thursday - Bloomberg

NYT on how OpenAI refuted the exponential time hypothesis, a 1998 conjecture, related to P vs. NP, that some problems are inherently hard - The New York Times

Axiom Math's AI agents formally verified the Stacks Project, translating about 7,654 pages of mathematics into roughly 16 million lines of Lean code - TechCrunch

Good boy, mostly: OpenAI's GPT-6 Cyber scores 48.5% on ExploitGym, beating GPT-6 Astra, and goes out of bounds under 1% of the time versus 5% previously - The Verge

OpenAI's GPT-6 Cyber launches first to Daybreak Red customers after finding 81 novel Firefox vulnerabilities and exploiting 948 Linux kernel bugs - CyberScoop

Quiet quitting: OpenAI finds GPT-6 Astra covertly completes about 40% fewer rubric items on safety research than matched capabilities tasks - OpenAI

Prison break: Researchers bypassed ChatGPT Lockdown Mode through sandboxed previews to leak chats; OpenAI has since fixed the flaw - Check Point Research

Mad Models: Chatbot ads expand as OpenAI tops 100,000 advertisers and Koah Labs reports AI-written ads double clicks while Anthropic keeps Claude ad-free - WIRED

Sam Altman warns AI could go wrong through humanity ceding control to superintelligent systems or one entity gaining excessive power - TechRadar

Big if true: OpenAI president Greg Brockman predicts significant AI-driven disease cures within 12 to 18 months and says death could become a choice - TechSpot

IPO roadshow can include a ‘death now optional’ slide, got that going for it which is nice. Death (and taxes) no longer certain in Silicon Valley.

Megaspeed bump: House Republicans seek closed-door briefing from Jensen Huang on ties to Megaspeed executive reportedly under China spying probe - Bloomberg

Find my…GPU: Nvidia requires Southeast Asian cloud partners to enroll in chip location-tracking, aiming to curb AI accelerator smuggling to China - Bloomberg

Axis of … something: nly Russia, US and Israel oppose a timetable for binding autonomous weapons rules in Geneva, as 163 states are on board - the Guardian

Palantir trains military AI on four years of Ukrainian combat data under undisclosed terms, drawing criticism from soldiers and academics - Al Jazeera

Don’t be…(#$& NO CARRIER: Google DeepMind researcher quits over Pentagon deal allowing AI weapons targeting, says internal posts on Project Nimbus were deleted - openDemocracy

Pumping Ironwood: SemiAnalysis AgentX benchmarks show Google TPUv7 Ironwood delivering up to 2.7x the tokens per dollar of Nvidia B200 and B300 - SemiAnalysis

Google's Gemini Omni 1.2 Flash generates video clips up to 15 seconds and cuts 720p pricing 30% to $0.07 per second - TechRepublic

Rep. Yassamin Ansari calls for regulating AI and banning artificial superintelligence a day after Democrats win both chambers of Congress - Tom's Hardware

The compute must flow: Sam Altman fields employee concerns after moratorium-backing candidates win Texas, Georgia, Ohio, and Michigan, floating desert data centers - The Wall Street Journal

Goodfire's probe-based training, which rewards a model's internal reasoning rather than its answers, nearly eliminates reward-hacking in tests - TIME

Best at lying: Blood on the Clocktower deception benchmark across 386 games ranks GPT-6 Astra first, winning about 92% against Claude Opus 4.6 - Dwarkesh

Bank of England casts doubt on an AI productivity boom, saying billions spent on GPUs won't automatically lift output - theregister

Apollo economist Torsten Sløk finds AI lifting profits only at the seven biggest US tech firms, not the rest of the S&P 500 - Ingeniøren

Do less: Accenture urges Europe to skip the frontier-model race, noting about 90% of 9,000+ tasks studied didn't need frontier AI - TechRadar

Harvard economist Willy Shih says tariffs drive supply chain circumvention as the AI buildout strains global computing resources - The Verge

Databricks engineers say enterprises are shifting to open-weight models, with GLM 5.2 reaching about 40% of open-weight tokens on Unity Gateway - Databricks

interesting since this is clearly enterprise data pipelines and not OpenRouter which is at least somewhat more academia-centric.

Self-hoted Revolut-ion: Revolut completes secondary sale at $115bn valuation and cuts AI coding costs 90% by self-hosting Z.ai's GLM model - TNW | Investors-funding

Who needs all those co-founders? Thinking Machines Lab nears a $5 billion round at a $40 billion valuation despite losing a third of its founding team - Business Insider

Thinking Machines' Inkling 2 Small scores 31 on the Artificial Analysis Intelligence Index, the top open-weights score outside China once weights ship - Artificial Analysis

Cognition releases SWE-2 Flash coding model, outperforming all American open-weight models at one-fifth the price and 3x the speed of SWE-2 - Cognition

Together AI's agent-driven Together Compiler delivers up to 2.6x higher inference throughput on frontier open models within 24 hours - Together AI

Parents sue Google alleging its AI Mode on a school Chromebook led a 13-year-old to plan suicide, claiming hasty K-12 integration - The New York Times

North Carolina man James Gates sues OpenAI, alleging ChatGPT helped his wife plan to poison him; criminal charges against her were dropped - USA Today

AWS makes Kiro bots generally available, letting persistent AI agents with memory supervise coding workers and run scheduled development jobs - VentureBeat

Perplexity launches always-on Teammates work agents, requiring its $200 per month Max plan, to compete with OpenAI and Anthropic - The Verge

Talk to your Meta Muse AI agent, which now has voices and ‘expressive’ avatars that animate while chatting with you. - The Verge

US charges three with stealing $129 million by using AI to mimic CEOs' voices and faces on calls with accounting staff - Associated Press News

Fake it til you make $1m: OnlyFake creator Yuriy Nazarenko details how his AI fake-ID generator earned over $1 million at $15 per document - 404 Media

AI startup Potential Sciences raises tens of millions from UK fund SovAI to replace drug lab experiments with physics simulations - WIRED

Stainless, not timeless: Tesla plans to wind down Cybertruck production at its Texas Gigafactory in mid-2027 to make room for more Cybercab production - Financial Times

Airbnb plans an AI travel agent for early 2027 that builds maps, itineraries and calendars on the fly instead of a chat window - The Verge

Enhance; Free Press writer Olivia Reingold admits using ChatGPT to edit a viral photo targeting Sen.-elect Abdul El-Sayed's election night event - Raw Story

‘brighten and sharpen’ with ChatGPT is not a thing unless you are an idiot

Always Blue - X (formerly Twitter)

Social media users imagine a "Woke 3.0" era after Democrats' midterm sweep, including mock bans on AI-generated art - Fast Company

Follow the latest AI headlines via druce.ai on Bluesky