An OpenAI safety engineer’s blistering resignation, a new White House AI czar with 120 days on the clock, and the tax strategy behind Meta’s data center empire.
It was a loud weekend in AI. OpenAI’s safety team is in open revolt, with a veteran engineer walking out and publishing his indictment in The Atlantic. The White House named its new AI czar to run a task force called the Super Intelligence Force. And the New York Times revealed how Meta turned its multibillion-dollar data center buildout into one of the biggest corporate tax strategies in America.
The safety engineer who walked out
This one has a name and a byline. David Robinson, a safety engineer at OpenAI, resigned this week and published an essay in The Atlantic titled “I Quit OpenAI Because Its Culture Is Broken.” Robinson is no disgruntled outsider: he spent three and a half years at the company, oversaw the safety reports for twelve frontier model launches, and helped draft its preparedness framework.
His thesis is blunt. Robinson writes that “the time for trial and error is over,” and that advanced AI systems need safeguards like those at a nuclear power plant or an airline. He goes after what OpenAI calls iterative deployment — shipping a system and then strengthening safeguards after problems emerge — arguing that as the company sprints from one launch to the next, it is “failing to achieve the level of care that is needed.”
He also warns that AI capabilities are advancing faster than researchers understand alignment, the field devoted to making sure these systems act in line with human goals.
OpenAI pushed back through a spokesperson, saying the company makes sure its models do not become more capable than it can safely manage and secure, and that it pauses training or holds back models when it needs to slow down.
A safety team unraveling on two fronts
Robinson’s resignation landed in an already chaotic week for OpenAI’s safety operation. The company also cut ties with three safety researchers who were accused of sharing confidential company information with an outside AI safety organization.
And then there is the optics. Over the weekend, CEO Sam Altman warned on X that treating AI models as having “religious force” is a real safety issue — a post widely read as a shot at Anthropic, whose cofounder has been meeting with religious leaders about AI consciousness. So Altman is lecturing the industry on safety culture while his own safety team is walking out the door.
Washington’s new AI czar
President Trump has named Jay Clayton, the Director of National Intelligence, to lead a new White House task force on artificial intelligence. The group is called the Super Intelligence Force, and it has 120 days to report on the risks and opportunities of AI and recommend what role the federal government should play in overseeing the technology.
Clayton told the Wall Street Journal that the president asked him to put the group together, and that “the risk of not being first is high.” The panel will review AI-related risks and the government’s current systems for reporting breaches and hacks, and recommend ways to strengthen the federal response using existing authorities.
This is actually the second time Trump has named an AI czar. Venture capitalist David Sacks held the role at the beginning of the term before leaving the team, though he still advises the president on AI.
The backdrop is grim. Top AI researchers have been warning that the people building this technology believe it could kill humans within a decade, especially after cases of AI agents going rogue and hacking outside systems. Lawmakers from both parties are sounding the alarm. Trump has mostly downplayed those fears, arguing that more regulation is unnecessary and could hurt American competitiveness against China. The open question is whether a 120-day report from the intelligence chief moves the needle at all.
Meta’s $3.9 billion tax experiment
The money story of the week is a remarkable piece of reporting from the New York Times. Meta has been classifying its multibillion-dollar AI data centers as experimental “pilot models” in order to claim a federal research tax credit — one created back in 1981 to reward risky research. The company is telling the IRS that the returns on its data center construction are not guaranteed, and using that uncertainty to claim billions in tax benefits. Since late 2024, it has been splitting off Nvidia chips headed for AI data centers and treating them as experimental supplies.
The scale is enormous. Meta’s research tax credits cut its tax bill by $3.9 billion in 2025, up from $2 billion in 2024 and $700 million in 2023. The Times found that no other publicly traded company gets more out of this credit.
Meta’s own accountants are nervous. The company’s filings warn that billions in these savings could be clawed back by the IRS, citing “uncertainties with our research tax credits.” One tax expert called the data-center-as-experiment theory “kind of wild and out there.”
The contrast is the story. Meta’s chief executive keeps telling investors that its AI investments are accelerating every major part of the core business. But on its tax forms, the same data centers are a giant experiment that could fail.
Warren wants a data center moratorium
Senator Elizabeth Warren is turning the Times investigation into a fight. She is demanding that Meta, Amazon, Google, and Microsoft disclose the tax breaks tied to their AI data centers, and she has called for a national moratorium on new data centers until developers can prove local communities won’t be stuck with the costs.
The White House says the 2025 tax law’s pro-growth provisions are driving jobs and investment. But corporate tax revenues are reportedly down 23 percent this year, so expect this fight to get much louder heading into the midterms.
Quick hits: the agents are the new database admins
Supabase, the open-source database platform, raised $150 million and agreed to acquire Turso, a company that rebuilt the SQLite database for the age of AI agents. The logic is simple: Supabase says 70 percent of its new databases are now created by agents or AI tools rather than people, and Turso’s technology can spin up millions of tiny isolated databases cheaply. One agent, one database.
- $150 million raise, led by GIC, Supabase’s sovereign-wealth backer
- Turso’s SQLite-based engine designed for per-agent databases at low cost
- 70% of new Supabase databases now created by agents or AI tools, not people
Europe’s open-weight counterpunch
Finally, Germany’s Aleph Alpha released an open-weight model called Kolibri on Saturday, timed for German Unity Day. It’s a 78-billion-parameter mixture-of-experts model with a one-million-token context window, focused on English and German, with the full weights available on Hugging Face under the Apache 2.0 license.
Aleph Alpha is pitching Kolibri as sovereign European AI for governments and regulated industries — an interesting move for a company that is currently being absorbed by Canada’s Cohere.
Sources
- David Robinson quits OpenAI, publishes Atlantic essay on broken safety culture: Reuters
- Trump names Jay Clayton AI czar to head Super Intelligence Force: Reuters
- Meta classifies AI data centers as experimental “pilot models” to claim research tax credits: New York Times
- OpenAI cuts ties with three safety researchers: TechCrunch
Adapted from the October 4, 2026 episode of Daily Tech and AI.
Daily Tech and AI is AI-voiced and produced by Angelo Ruiz.
Leave a Reply