Two of four frontier AI labs tell UK MPs they want binding rules
OpenAI and Anthropic told UK MPs that voluntary pledges no longer suffice, while Google DeepMind and Meta declined to back legislation. Teams building on these models now have vendors on the record disagreeing about whether they should be bound.
The Product Desk
OpenAI backed a statutory framework, starting with incident reporting. Anthropic: the rules should be binding. Google DeepMind: focus on standards. Meta pointed to Britain's existing regulators. AISI's director: AISI cannot prevent release of a model it judges too dangerous.
- Who has to decide decision OpenAI Backed a statutory framework; some elements should be mandatory, starting with incident reporting, claim 5
- Who has to decide decision Anthropic Liam Booth-Smith, on a statutory framework: "The rules should be binding.", claim 4
- Who has to decide decision Google DeepMind Owen Larter, on a statutory framework: "Focus on the standards is what we would be doing.", claim 6
- Who has to decide decision Meta Rob Sherman pointed to Britain's existing regulators: "not writing on a blank page.", claim 7
- What it limits constraint AI Security Institute (AISI) Director Henry de Zoete: no ability to prevent release of a model AISI judges too dangerous, claim 12
| Who | How | Kind | Claim |
|---|---|---|---|
| OpenAI | Backed a statutory framework; some elements should be mandatory, starting with incident reporting | decision | 5 |
| Anthropic | Liam Booth-Smith, on a statutory framework: "The rules should be binding." | decision | 4 |
| Google DeepMind | Owen Larter, on a statutory framework: "Focus on the standards is what we would be doing." | decision | 6 |
| Meta | Rob Sherman pointed to Britain's existing regulators: "not writing on a blank page." | decision | 7 |
| AI Security Institute (AISI) | Director Henry de Zoete: no ability to prevent release of a model AISI judges too dangerous | constraint | 12 |
What happened
- The Business, Innovation, Science, and Trade Committee heard the four firms on Tuesday, alongside four former staff turned whistleblowers and senior British civil servants.
- OpenAI's Tom Duff Gordon backed a statutory framework and an international convention, with incident reporting as the first element he wanted made mandatory.
- Cat Little, the Cabinet Office's top civil servant, confirmed to the committee that the UK's AI safety regime is voluntary at the moment.
- AISI director Henry de Zoete acknowledged the institute has no ability to stop a company releasing a model that AISI has judged too dangerous.
- OpenAI said it gives its models to the US body CAISI before deployment and will keep doing so for AISI only as long as the US government allows.
Why it matters
- Who has to decide decision The only starting point any vendor named was incident reporting, so it is the clause a buyer can request in a contract today without waiting for a statute.
- What it limits constraint UK pre-release testing happens only when companies hand models over, and OpenAI ties its sharing to what the US government permits, so a government test cannot be treated as a gate on any model in your stack.
- What doesn't add up contradiction With two firms for binding rules and two declining, a pitch that says the labs welcome regulation is true of only half the vendors in the room.
Next story · Science · Oct 5
Four AI labs raise no hands when New York's council asks who insures against catastropheConfirmed2 publishers
Asked directly whether he supported a statutory framework, Google DeepMind's Owen Larter said: "Focus on the standards is what we would be doing." [6] Meta's Rob Sherman pointed to Britain's existing regulators and told MPs they were "not writing on a blank page." [7] Anthropic's Liam Booth-Smith, asked whether his company's view was that AI safety now needs a statutory framework, said: "The rules should be binding." [4]
Committee chair Liam Byrne told WIRED the session was meant "to light up the wide differences in opinions between some of the wealthiest firms on Earth." [8]
AISI tests frontier models before release only when companies choose to hand them over. [9] Its director, Henry de Zoete, said the institute relies entirely on the cooperation of companies and foreign governments for access. [10] "What we're not doing is kite marking," de Zoete said. "We're not saying yes or no to any sort of model." [13]
De Zoete also put the limit in terms of geography. He said the companies building at the frontier are either American or Chinese and added: "There's only so much we, as the UK, can do in that kind of context." [14]
No bill, draft or date has been reported, and nobody has said who would receive incident reports or what would count as an incident. On this record a UK duty is a request from two of the four companies [16], and we think the vendor split is the part to plan around now.
For a team with more than one frontier vendor, we'd sort each vendor on two tests. The first is whether the vendor has said publicly that it wants binding rules: OpenAI and Anthropic have [1], and Google DeepMind and Meta declined to back legislation [2]. The second is whether your contract already obliges the vendor to tell you about safety incidents.
- Vendor backs rules, contract has the clause: the two agree and nothing needs changing. - Vendor backs rules, contract is silent: ask for the clause, since the vendor has said in public that it wants binding rules. - Vendor declined, contract has the clause: keep it. - Vendor declined, contract is silent: the vendor has not asked for a rule and your paper gives you no notice, so you learn of an incident when the vendor chooses to say so.
What to watch
- Whether the committee or the government puts a draft bill or a mandatory incident-reporting proposal on paper, including who would receive the reports and what counts as an incident.
- Whether Google DeepMind or Meta changes its answer on statutory backing after this session.
- Whether OpenAI keeps giving models to AISI before deployment, given its condition that the US government allows it.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence78
- Adoption
- Not scored
- Hype gap−5
- Incentives
- Not scored
- Confidence75
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Representatives for OpenAI and Anthropic told a British parliamentary hearing that voluntary pledges were no longer enough to ensure safety and that new laws should be introduced.
- [2]
Google DeepMind and Meta both declined to back new legislation.
- [3]
The Business, Innovation, Science, and Trade Committee, a cross-party panel of MPs, held an evidence session on Tuesday with appearances from four frontier AI firms, four former staff of the firms turned whistleblowers and senior British civil servants.
- [4]
Anthropic's head of Europe Liam Booth-Smith, asked whether the company's view was that AI safety now needs a statutory framework, said: "The rules should be binding."
- [5]
OpenAI vice president Tom Duff Gordon backed a statutory framework and an international convention, and said "there should be some elements that should be mandatory," starting with incident reporting.
- [6]
Asked directly whether he supported a statutory framework, Google DeepMind's Owen Larter said: "Focus on the standards is what we would be doing."
- [7]
Meta's Rob Sherman pointed to Britain's existing regulators, telling MPs they were "not writing on a blank page."
- [8]
Committee chair Liam Byrne told WIRED after the hearing that the committee set out "to light up the wide differences in opinions between some of the wealthiest firms on Earth."
- [9]
Britain's AI Security Institute (AISI) tests frontier models before release, but only when companies choose to hand them over.
- [10]
Henry de Zoete, director of AISI, acknowledged that the institute was entirely reliant on the cooperation of the companies and foreign governments in order to gain access.
- [11]
Cat Little, the top civil servant at the Cabinet Office, which houses AISI, was pressed on whether the whole regime was voluntary and said: "At the moment, yes, it is."
- [12]
Asked what would happen if AISI judged a model too dangerous but a company released it anyway, de Zoete acknowledged AISI had no ability to prevent the release.
- [13]
De Zoete said: "What we're not doing is kite marking. We're not saying yes or no to any sort of model."
- [14]
De Zoete said the companies building at the frontier are either American or Chinese and added: "There's only so much we, as the UK, can do in that kind of context."
- [15]
Duff Gordon said OpenAI provides its models pre-deployment to CAISI, AISI's US counterpart, and will continue to provide them pre-deployment to AISI in the UK as long as it is able to do so under the US government.
- [16]
Two of the four frontier AI firms at the hearing backed binding rules and two declined to back legislation.
Sources
2 independent publishers whose own reporting we read for this story.
- dev.toOpenAI and Anthropic call for new laws on artificial intelligence
1 article · October 13, 2026
- wired.comOpenAI and Anthropic Tell British Lawmakers: Regulate Us
1 article · October 13, 2026
Topics and entities
Entities
- OpenAI
- Center for AI Standards and Innovation (CAISI)
- Google DeepMind
- Meta
- Business, Innovation, Science and Trade Committee
- Anthropic
- UK AI Security Institute