Thursday, September 10, 2026
Today's stack is a bit heavier than usual: a safety researcher just walked out of Anthropic warning that things are moving too fast, while a longtime AI worrier just walked into OpenAI's boardroom. Grab your coffee, this one's worth reading slowly.
An AI Safety Researcher Just Quit, Saying We're Gambling With Our Lives
Jacob Coxon, a safety researcher at Anthropic, publicly resigned this week warning that AI systems capable of improving themselves are becoming too dangerous to control. He compared the race inside AI labs to a mini nuclear weapons project moving forward without enough safety brakes, and called for competing companies to sign pacing agreements, meaning they'd agree to slow down together instead of racing each other. Just hours later, another Anthropic researcher said there's more than a 10 percent chance AI could kill all humans by the end of the decade. Both work at one of the industry's most safety focused labs, which makes the warnings harder to wave off.
What this means for you: Even the people building these tools are getting nervous. It's a reasonable moment to stay curious about AI but a little more careful about how much control you hand it.
What this means for your business: If safety experts inside the labs think things are moving too fast, expect regulators and insurers to start asking harder questions soon. Build human checks into any high stakes AI decision now, before it becomes a legal requirement.
Source: TechCrunch
Radar 01
A Longtime AI Worrier Just Got a Seat on OpenAI's Board
Paul Christiano, a researcher who has spent years warning that AI companies move too fast without enough safety testing, is joining OpenAI's nonprofit foundation board and its safety committee. He previously ran a research group focused entirely on keeping AI systems aligned with human intentions. His appointment lands right as OpenAI faces mounting pressure over safety incidents tied to its newest models.
What this means for you: Someone known for sounding alarms about AI risk now has real influence inside one of the industry's biggest labs, a sign these worries are reaching boardrooms, not just research papers.
What this means for your business: Expect OpenAI's policies to lean more cautious, with more safety reviews and possibly slower feature rollouts. If your company depends heavily on OpenAI's models, plan for that shift.
Source: TechCrunch
Radar 02
Companies Are Spending Less on AI Per Employee, and Nobody's Sure Why
New data shows spending on AI per employee actually dropped at major companies in August, even as AI adoption headlines keep piling up. Falling costs for AI tokens, the small chunks of text these systems process, and cheaper models mean companies get more AI use for less money. But it also raises a tougher question: are employees actually using AI more, or has the early rush of enthusiasm quietly cooled off.
What this means for you: If your own AI habits have plateaued lately, you're not alone. Worth checking whether your team's early excitement faded into occasional, half hearted use.
What this means for your business: Falling AI costs help your budget, but flat usage per employee is a warning sign for adoption strategy. Check whether your rollout gave people real daily reasons to use AI, not just access to it.
Source: TechCrunch
Radar 03
Google Built a Map of Every Possible DNA Typo
Google DeepMind released a giant map predicting what happens if any single letter in human DNA changes, covering nine billion possible variants. Think of DNA as an extremely long instruction manual for the body. This tool predicts, letter by letter, what a typo in that manual might do. Scientists can use it to figure out which genetic changes are likely harmless and which might cause disease, potentially speeding up research into new treatments.
What this means for you: This is the kind of quiet AI breakthrough that could eventually speed up diagnosis and treatment for genetic diseases affecting you or people you know.
What this means for your business: If you work in health care, biotech, or pharma, tools like this could shave years off early stage research. Worth tracking how partners and vendors start building on top of it.
Source: Google DeepMind Blog
Try This Today
Ask your team lead one simple question today: for our most important AI powered decision, is there always a human who checks it before it ships? If you're not sure, that's this week's fix.
Quick Hits
- US officials are urging American AI companies to identify Chinese users and quietly give them weaker, less capable versions of their models, amid growing accusations that Chinese AI firms have been aggressively copying US systems. [1]
- The Justice Department is investigating whether Nvidia structured its 20 billion dollar licensing deal with AI chip startup Groq specifically to dodge antitrust review, adding fresh scrutiny to Nvidia's growing web of AI chip partnerships. [2]
- New research proposes a way to measure whether giving AI agents long term memory, meaning the ability to remember past interactions, actually helps them finish real tasks, or whether the extra cost of maintaining that memory outweighs the benefit for many business use cases. [3]
Enjoyed today's read? Help me get this in front of more people like you.
Start every morning a little sharper.
Free. Takes 5 minutes to read. No spam.
