Good morning. It's Friday, October 9, and we're covering OpenAI’s safety culture, Google’s push for persistent AI coworkers, and the debate over AI-generated mathematical research.
Plus: yesterday’s poll results on AI in government, a Friday Fact about a planet with a day longer than its year, and an essay on why every AI company wants to be your assistant.
OpenAI released more than 700 AI-written math manuscripts, but most have not gone through the independent review that gives new results credibility. Only a small share include the model’s reasoning, and many lack a formal proof in Lean, a language that lets a computer check each logical step.
Anthropic updated its usage policy to spell out bans on deceptive political campaigns, election interference, weapons work, surveillance and attempts to misuse or weaken its models. The changes take effect November 12 and give customers a clearer list of behaviors that can lead to enforcement.
President Trump said the White House considers people who use “artificial intelligence” instead of his preferred term, “super intelligence,” to be enemies. The statement follows an executive order directing federal agencies to use the new name, even though the technology industry still overwhelmingly calls it AI.
OpenAI’s annualized revenue is reportedly approaching $50 billion, about $20 billion below a figure previously circulated among investors. The difference may partly reflect how the company counts revenue, but it raises the stakes as OpenAI spends heavily on computing and prepares for a possible IPO.
FRIDAY FACT
Which Planet Has a Day Longer Than Its Year?
On one planet in our solar system, a single rotation takes longer than an entire trip around the Sun. Even stranger, the Sun rises in the west. The answer is at the bottom of the newsletter.
POWERED BY CODERABBIT
Prioritize Your PR Queue
Review the PR that matters most. CodeRabbit Triage ranks work across repositories, tells you what to review next and what needs your action. It also flags low-risk PRs that may be ready to close, so your queue shrinks effectively.
ORIGINAL
Why Every AI Company Wants to Be Your Assistant
The model race is becoming an assistant race. OpenAI, Google, Microsoft, Anthropic, Meta and SpaceXAI are all building systems meant to become the first place people go to start work—not just a chatbot they visit for an answer.
That position becomes more valuable as an assistant learns a person’s files, coworkers, customers, preferences and history. The same context that makes the product useful also makes it harder to leave, creating a new software lock-in around the AI that knows how you work. → Read the full article here.
AI SAFETY
Why OpenAI’s Former Safety Lead Quit
David Robinson resigned from OpenAI after leading the reports that explain the safety risks of new models. He told Ezra Klein that OpenAI and the rest of the industry are now adding new reasoning, tools and coding abilities so quickly that major changes no longer arrive only with a brand-new model.
Robinson’s concern is less about one dramatic failure than a culture that treats safety as something to add after the race has started. He argues that labs are building guardrails around systems trained to find ways around obstacles, and he no longer believes the industry has the habits needed to manage that risk. Watch whether his departure changes how regulators judge the company’s own safety claims. → Read the full article here. (Paywall)
WORKPLACE AI
Google Wants Gemini to Follow Your Work Everywhere
Google introduced a persistent Gemini agent for businesses that can research, create content, write code and keep working for hours or days after a user closes a laptop. It carries the same memory across devices and apps, can work inside Google Workspace and Microsoft 365, and can create smaller agents for parts of a job.
The larger bet is that companies will treat an AI agent like a coworker with its own identity, email address, storage, permissions and audit trail. Google says Gemini can choose among its own models and Anthropic’s Claude while keeping the company’s data and context in one place. Watch whether businesses accept that access controls and responsibility for mistakes become real operating questions. → Read the full article here.
NEWS
What Else is Happening
Arena Raises $200 Million at a $3.1 Billion Valuation: AI evaluation firm Arena raised $200M after reaching a $100M annualized revenue rate.
Energy Department Selects Four Robotics Projects: Four national labs won up to $30M to develop reusable robotics for scientific research.
Manus Raises $500M After Meta Split: Chinese AI-agent firm Manus raised over $500M in its first round since its planned Meta deal fell apart.
Goodfire Tests Monitors That Watch Inside AI Models: Its activation-based monitors added under 2% latency while detecting harmful agent behavior.
Amazon Builds Backup Power for Data Centers: AWS created Project Falcon to let Mississippi data centers switch to backup power during peak demand.
FRIDAY FACT
A Day on Venus Is Longer Than a Year
Venus takes just 225 Earth days to orbit the Sun, but 243 Earth days to complete a single rotation. That means an entire Venusian year passes before the planet finishes turning once on its axis. Even stranger, Venus spins backward compared to most planets, so the Sun rises in the west.
There's another twist: because Venus rotates opposite to its orbit, the time between sunrises is only 117 Earth days. And despite being farther from the Sun than Mercury, Venus is the hottest planet in our solar system, with surface temperatures reaching 870°F (465°C), hot enough to melt lead.
POLL RESULTS
Do You Trust AI in Government?
Here's how you voted: AI as an advisor led with 53%, while 24% said AI should make government decisions and 13% preferred humans only.
🟨🟨🟨⬜️⬜️⬜️ Decision-maker — AI could make better decisions. (24%)
🟩🟩🟩🟩🟩🟩 Advisor — AI should only inform decisions. (53%)
🟨⬜️⬜️⬜️⬜️⬜️ Human-only — Decisions should stay with people. (13%)
🟨⬜️⬜️⬜️⬜️⬜️ Not sure — It's too early to tell. (10%)
