Notes from the Intersection No. 008
Carmen Onchain · 16 October 2026 · 7 min read
When AI's loudest voices ask to be regulated
Something unusual happened in Westminster last week. Two of the loudest companies in artificial intelligence stood in front of British lawmakers and asked, in plain language, to be regulated. OpenAI and Anthropic made the ask, Wired carried the story on October 13, and the framing has been echoing through policy circles that only three weeks ago were still arguing about whether AI needed any rules at all.
For most of the last decade the loudest ask from the frontier labs has been the opposite of that. Give us room. Do not slow us down. The pivot to please slow us down is not a philosophical conversion. It is a reading of where this is going by the companies that stand to lose the most if the reading is wrong.
What concerns me is how quickly the frame has moved and how little the move has been examined. Three weeks ago the public policy conversation was still debating whether AI needed rules at all. Last week two of the biggest labs told a parliament they want them. That is not a small thing, and not obviously a good one. A rule written by a lab tends to protect the lab. A rule written by a regulator who sat with the users, workers, and small operators harmed when an agent misfires in a real business tends to protect them first.
The real question is not whether to welcome the shift. It is to notice who is writing which rules, and for whom.
Deep Dive
What to make of the regulate-us moment
What is actually happening here
The headline from Wired on October 13 was blunt. OpenAI and Anthropic told British lawmakers to regulate them. The language underneath would have been unthinkable in a frontier-lab filing eighteen months ago. The companies said voluntary commitments are not enough. They asked for rules.
In a different vocabulary the same week, Microsoft's leadership repeated something it has been saying publicly for a year. A carton of milk, in that framing, faces more safety requirements than the frontier AI model running on Microsoft's own cloud. The line became a headline because it is embarrassing to be true.
Why this matters beyond the money
It would be easy to read all of this as a lobbying story. It is a lobbying story. It is also a legitimacy story, which matters more.
For most of the past decade, the AI industry has argued the technology is too novel to regulate well. The argument rhymes with the one before it. In 2016 it was crypto. In 2020 it was social media. In each case the too-novel-to-regulate argument bought another five years of unregulated growth and made the eventual regulation more punitive than it would have been if the industry had helped write it. The labs are watching that history and placing a smarter bet. Instead of arguing for no rules, argue for the rules that let the current winners keep winning.
Fortune's AI editor Jeremy Kahn made the sharper counterargument on October 13. In his column on strict liability, he said the standard "is a way to try to incentivize AI companies to adopt a safety culture and exercise more care in the absence of industry consensus and standards about what best safety and security practices look like." Strict liability means a company is on the hook for harm its product caused whether or not it was negligent. Nuclear plant operators live under it. Aircraft makers live under it. If it lands on AI, the labs will owe for harm their models cause, with or without a defense.
That is the stick the regulate-us ask is quietly trying to bend. If the labs agree on a lighter federal standard with a regulator in the room, the state courts have less work to do. The question that follows is who writes those federal rules alongside the labs. If the rulemaking happens in private rooms with the biggest firms, the smallest operators lose. If it happens in public rooms with the users, workers, and small operators an AI tool is most likely to harm first, the solo founder with a chatbot in production gets the defense she needs when it misfires on a paying customer.
Blockchain enters this conversation in a specific and often misunderstood way. A tamper-evident log, which is a fancy way of saying a record the agent cannot rewrite, is the kind of mundane infrastructure a regulator points to when asking how you know what the model actually did on a given day. A decentralized identity layer, a way of signing an action so anyone can verify who authorized it, is the kind of infrastructure a regulator points to when asking who signed what, and when. These are not science-fiction primitives. They are shipping. They are the audit trail strict liability will eventually require of anyone selling a model into a regulated workflow.
What builders and operators should be watching
If you run a business that uses AI tools in 2026, the regulate-us moment is a signal to run a short exercise on your own stack. Pick any AI tool you depend on and write down the answer to a single question. Who is accountable when the tool misfires on one of your customers?
Most teams find, if they are honest, that the real answer is nobody with real exposure. The vendor's terms of service say the vendor is not liable. The integrator's contract says the integrator is not liable. You carry the customer, and your defense is whatever log you kept. That is the gap strict liability is meant to close, and it is the gap the regulate-us ask will try to narrow first.
The next twelve months will likely tighten the gap in two ways. Strict liability conversations like the Fortune piece this week will move from academic circles into state statutes, and the first jurisdiction to pass a bill will set the template others copy. The frontier labs will likely sign onto softer federal rules first, which the biggest firms will absorb and the smallest will struggle to implement. The practical steps have not changed, but the timing has become urgent. Keep a clean record of which model made which decision and when. Keep a human in the seat where the decisions matter. Pay for an audit trail you own, not one the vendor owns. Know who you would call if the tool harmed a customer, and keep that name in your phone before you need it.
Strict liability is not here yet. The UK ask is not law yet. The question for the operator is not what will be legal in two years. It is what you can afford to lose while the lawyers argue about where the line should fall.
Microsoft said it in a sentence that cannot be unheard. If a carton of milk gets more scrutiny than the model running a bank's compliance desk, the rulebook is not incomplete, it is upside down. The line is doing political work two years of policy papers could not, because every parent reaches for the milk analogy and every regulator now has to answer for it. Watch which legislators repeat it this quarter.
IJR · 12 October 2026
The real stick in the AI debate is not a hearing in Westminster, it is a jury in a state court somewhere, and the labs are reading the room earlier than they used to.
Fortune · 13 October 2026
The weak link in any stack is rarely the vendor with the biggest logo, it is the box between you and that vendor, and this week it drained $90 million from people who thought the brand was the only check they needed.
Quartz · 14 October 2026
The convergence of blockchain and AI is starting in the plumbing, where an autonomous agent's intent on a public ledger is a target the moment it is readable, and encrypting the waiting room is a cheap, boring, correct answer.
Biztoc · 12 October 2026
When your AI tooling layer is a single-vendor dependency, the only responsible move for an operator is to know what the alternative actually ships, and this week the alternative shipped enough to be worth a look.
Webpronews · 13 October 2026
The regulate-us moment is useful only if we notice what it changes. For the first time in a decade, the biggest labs are reading the political weather and asking for a rulebook. If we leave them alone in the room, the rulebook will look like them. If we fill the room with the users, the workers, and the small operators that strict liability is designed to protect, the rulebook will look like the business you are actually trying to run.
The encouraging part is that the audit trail that rulebook will eventually ask for is one blockchain quietly knows how to build. A tamper-evident record, a decentralized identity, an encrypted waiting room for the agent's trade. These are mundane primitives, and they are shipping. Pick one AI tool in your business and write down who you would call when it misfires. Start there.
What does "regulate us" actually mean when OpenAI and Anthropic say it?
In October 2026, OpenAI and Anthropic told British lawmakers that voluntary commitments are not enough and the industry would benefit from binding rules. In practice they are asking for a federal or supranational rulebook that applies to all frontier labs, which they prefer to a patchwork of state-level laws decided by juries one lawsuit at a time.
What is strict liability for AI companies?
Strict liability means a company is on the hook for harm its product caused, whether or not it was negligent. Nuclear plant operators and aircraft makers already face this standard. If it is extended to AI, labs would owe damages when their models cause harm regardless of what safety steps they claim to have taken.
How does blockchain fit into AI governance?
Blockchain's role in AI governance is less about replacing regulators and more about giving them a record they can audit. Tamper-evident logs let a lab prove or disprove what a model did on a given day. Decentralized identity standards let a regulator confirm which agent signed which action. These primitives make strict liability enforceable, not theoretical. This guide on blockchain and AI convergence walks through the architecture choices.
What should a small business do while AI regulation is being decided?
Keep a clean record of which model made which decision and when, keep a human in the seat where the decisions matter, and know who you would call if the tool harmed a customer. This guide on who governs AI is a good starting point. Do not wait for legislation. The first jurisdiction to pass a strict liability rule will set the template.