OpenAI’s safety bars ask, Uber’s $100M into Atoms, and Anthropic’s settlement fights

In this Brief

• OpenAI An Alien Mind: voluntary slowdowns + mandated safety bars

• OpenAI research intern: 3.1 agent-workdays; $600 / $7,000 token days

• Uber $100M into Atoms / Levandowski; company denies robotaxi

• Campaigns: 39 candidates on OpenAI seats; RNC ~$9,700

• Anthropic $1.5B settlement: authors vs publishers and agents

OpenAI An Alien Mind

Setup: Jakub Pachocki at OpenAI published An Alien Mind on Sunday arguing no lab has solved alignment and monitoring, getting models to pursue intended goals and watching what they do, well enough to keep scaling at maximum speed for much longer. He wants the Preparedness Framework, OpenAI's internal safety rules, evolved into widely mandated safety bars enforced by outside auditors or governments, and he hopes voluntary slowdowns become commonplace. Business Insider notes Altman reposted it as important. Essay, not a pause order.

Take: The lab that shipped Astra, its latest model, is writing the caution. Mandated bars are the ask. Slowdowns are the hope.

Longer cut: OpenAI's chief scientist says no lab is ready to keep scaling at full speed.

Live X: Aaron_Harme · site promo.

OpenAI research intern scoreboard

Setup: OpenAI's research-acceleration post says the company hit its September 2026 automated research intern goal. As of mid-August: 3.1 agent-workdays per human workday (eight hours of model labor on the company clock).

Median researcher (by agent usage) spends >$600/day on inference, the cost of running the models, at API prices. The 90th percentile research-org user burns >$7,000/day.

Take: March 2028 goal remains an automated AI researcher. People still set priorities and decide scale, pause, or deploy. Scoreboard up. Full recursive self-improvement, machines increasingly driving their own development, still not solved on the company's own page.

Live X: Aaron_Harme.

Uber $100M into Atoms

Setup: The Next Web, citing the Financial Times, writes that Uber put $100 million into Atoms, Travis Kalanick's venture, after Atoms hired Anthony Levandowski. Sources say Atoms is developing robotaxi tech, cars that drive paying passengers without a human driver, with talks about running it on Uber's network. Atoms called itself industrial software with no plans to enter that market.

Uber declined to comment. The check size is not disputed.

Take: Denial on record. Sources still say robotaxi. The $100 million is not in dispute.

Longer cut: Uber put $100 million into Travis Kalanick's Atoms after Atoms hired Anthony Levandowski.

Live X: Aaron_Harme · site promo.

Campaigns buy ChatGPT seats

Setup: Slashdot, paraphrasing the Washington Post, writes that 39 congressional candidates reported paying for an OpenAI subscription this cycle. Around 30 political action committees (PACs) and parties reported payments. The RNC was the largest political spender at roughly $9,700.

Two candidates disclosed advertising use against company policy. Post tests found ChatGPT inconsistently enforcing bans on targeted fundraising copy.

Take: Paid seats sit under soft bans. Same prompt, different answer.

Longer cut: Campaigns are buying ChatGPT seats while OpenAI's election bans slip under the same prompts.

Live X: Aaron_Harme · site promo.

Anthropic settlement allocation fights

Setup: Anthony Ha at TechCrunch writes that authors are fighting publishers, and sometimes agents, over cuts of Anthropic's $1.5 billion piracy settlement after a judge said training on copyrighted material can be fair use but pirating was not. Nearly 500,000 titles; up to $3,000 per work. In-print traditional: 50/50. Self-published or rights-reverted (author got the copyright back): author 100%.

Writers Beware's Victoria Strauss logs publishers claiming works they no longer control and seeking 100% when entitled to 50%. Some literary agents are filing claims too.

Take: Three thousand a book. The cut is the fight. Allocation disputes, not a new finding that publishers cheated.

Live X: Aaron_Harme.

Also noted

TNW / NYT reports that Kenya essay mills that once paid ≥40,000 people in Nairobi collapsed after ChatGPT. Live X: Aaron_Harme.

TNW / Information writes that Thinking Machines talks of a $5 to $6B raise with Nvidia roughly half, not closed. Live X: Aaron_Harme.

Gizmodo reports AMC's CEO is furious at Robinhood for tokenizing the stock without permission.

TNW writes that Isar Spectrum reached orbit on its second flight, a continental Europe commercial first.

Gizmodo reports Tesla's Cybercab first-responder guide may require push and hand-turn tires.

The Sunday Brief covered SB 813 / METR $400k, Copilot discovery, Anthropic IPO mid-Oct, Berlin Rhysida, and CLARITY.

Sources

OpenAI — An Alien Mind; research acceleration

Business Insider — Altman repost color

TNW / FT — Atoms / Uber $100M

Slashdot / Washington Post — campaigns OpenAI seats

TechCrunch (Ha) — Anthropic settlement allocation

TNW / NYT — Kenya essay mills

TNW / Information — Thinking Machines talks

Gizmodo — AMC tokens; Cybercab

TNW — Isar Spectrum

Live BF / X links as above (already-live only; not Mon morning queue)