Brockman pulled a $25M super PAC pledge as AI researchers organized

Share
Brockman pulled a $25M super PAC pledge as AI researchers organized

The self-regulation pledge six CEOs signed this week has a workforce problem attached: at the frontier labs, the researchers are organizing — and one president just moved money because of it.


Axios reports a quiet rebellion inside the frontier labs, and its hardest news peg is that OpenAI president Greg Brockman scrapped a pledged second $25 million donation to the pro-AI super PAC Leading the Future. Axios' account — built from interviews with researchers inside OpenAI, Anthropic and Google DeepMind — describes a small cadre of elite researchers who now wield unusual influence over their own companies' policy positions: OpenAI backed an Illinois transparency bill and, at the last minute, a California independent auditing bill, and an OpenAI investor told the outlet that AI CEOs have to embrace regulation to keep their best research talent. The Wall Street Journal independently confirms the money story: Brockman told employees in June that the second $25 million pledge was off, after a first $25 million went in last September. The petition campaign Pacing the Frontier has drawn more than 1,000 verified signatures from corporate email addresses at the major labs, and compensation for elite researchers has reportedly climbed into the hundreds of millions of dollars — talent this expensive gets a say. OpenAI's line is that "what ultimately drives our policy positions is what we're seeing in our own technology," but one Washington observer quoted by Axios says the internal pushback has left the company "no longer seen as or acting as a leader in driving policy." Read against this week's firings — OpenAI fires three safety researchers over outside info sharing — the pattern is the same story from both ends: labs trying to control information outward while their own people push policy positions inward. The money is the tell. When the president of the biggest lab starts unwinding political donations, the rebellion is no longer a mood — it's a cost.


Allen AI open-sources AstaBrief 8B, the model that powers Fast mode in its scientific report generator Asta. The 8-billion-parameter model — a fine-tune of Qwen3-8B, Apache 2.0 licensed, weights and training data both released — writes a full cited report in one pass instead of section by section, cutting Asta's average report time from 178.5 seconds to 51.1 seconds, about 3.5 times faster, while staying competitive with the Claude-powered pipeline it was measured against. Ai2 is candid about the limits: most of the training and evaluation happened in 2025, and the team has not rerun it against today's frontier models, so the release is best read as a validated recipe — 90,000 filtered real research queries turned into 47,000 supervised examples plus 6,000 preference pairs — rather than a state-of-the-art claim. The point that matters for anyone who can't ship research queries to a proprietary API: open weights mean hospitals, labs and universities can run cited report generation behind their own firewall.


Slovenia's .si domain registered about 44,000 names in September — up from roughly 2,000 in August — after Trump's executive order renamed "artificial intelligence" to "super intelligence" across the US government. Registry SI told the BBC it saw a 2,199% jump, with 11,000 new addresses recorded on September 30 alone, the day after the order signed; hosting platform Hostinger says .si is now its second most popular extension after .com, with more than half of buyers coming from the US and India. The echo of Anguilla's .ai windfall is obvious but the arithmetic isn't: .si costs around $12 a year against $90 for .ai, and the money goes to the registry rather than a national treasury — as Hostinger's domain chief put it, for Anguilla .ai became a second economy, for Slovenia .si is a nice bonus. Only about 3% of the .si registrations on Hostinger are explicitly AI-related, which suggests most of this is speculation rather than use. We covered the pledge that started the week — Trump's AI rebrand lands with a six-CEO self-audit pledge.

What to watch: whether more money follows Brockman's reversal out of the pro-AI super PACs. Should frontier-lab employees be able to push their own companies' policy positions in public? Tell us in the comments.

Read more

OpenAI drops 372 math results, nearly all from one prompt

OpenAI drops 372 math results, nearly all from one prompt

A dense news day across three fronts: OpenAI published a mountain of mathematics from an unreleased model, SpaceX reportedly wants the debt market to fund its Nvidia orders, and Anthropic restructured its cyber program around three tiers of trust. OpenAI published 372 mathematical result families from an internal frontier model — and says nearly all of them came from a single prompt handed to a single AI agent. The release comprises 722 manuscripts touching hundreds of open questions, plus Lea

Underdog launches a private on-device AI assistant, backed by a16z

Underdog launches a private on-device AI assistant, backed by a16z

The privacy split in consumer AI got a new entrant tonight, GitHub turned code-review benchmarks into a vendor-bias argument, and OpenAI opened its oddest API to everyone. Underdog launched in invite-only beta: an on-device AI assistant that keeps your data on your machine and never charges you a subscription. Self-taught coder and Thiel Fellow Sigil Wen — who moved to Silicon Valley at 17 and lived in an AI hacker house with Andrej Karpathy — built his own inference engine, Husky, to run a 2

Today in AI — October 6, 2026

Today in AI — October 6, 2026

A day of second-order moves: the labs are buying task data from software vendors instead of scraping the web, Waymo is borrowing to fund the robot world, and the biggest bank in the US just put a number on what Anthropic's latest model cost it in risk. Models & Research * OpenAI is training GPT-6 Astra on Ironclad's real contracting work. Ironclad staff helped turn 11 tasks across legal, commercial and procurement work — setting up NDAs, approval workflows, clauses that change by jurisdict

Meta, Walmart and Stripe publish the Personal Agent Protocol

Meta, Walmart and Stripe publish the Personal Agent Protocol

The agent economy is writing its rulebook tonight: one open standard for AI bots at the checkout, a cheaper image model from Google, and Anthropic turning bug-hunting into a tiered product. Meta, Walmart, Stripe and Sierra are publishing an open "personal agent protocol" — a standard that defines how personal AI agents interact with businesses online. The group behind it reads like a cross-section of agentic commerce: Meta, Sierra, Walmart, Stripe, Shopify, Genesys, Rocket, NiCE, Decagon and I