UK Artificial Superintelligence Bill (ControlAI draft, May 28, 2026; introduced September 8, 2026)

The Artificial Superintelligence Bill is a UK bill drafted by ControlAI (authored by Tolga Bilge and Andrea Miotti) that would criminalize the development, deployment, or operation of artificial superintelligence inside the UK, impose personal liability on senior decision-makers responsible for prohibited activity, and oblige the UK to pursue an international treaty banning superintelligence worldwide. Published May 28, 2026, it proposes to prohibit the development of artificial superintelligence outright rather than to regulate or audit it. No other major-jurisdiction legislative text currently asserts that foreclosure position in operative statutory language; the EU AI Act and similar regimes impose obligations on frontier capability without foreclosing it.

Two names. ControlAI and its campaign materials call the instrument the Artificial Superintelligence Security Bill. The bill record maintained by the UK Parliament, at bills.parliament.uk/bills/4288, titles it the Artificial Superintelligence Bill, without "Security" (Source: bills.parliament.uk). The bill text itself resolves which is the instrument's own name: section 44 provides that "[t]his Act may be cited as the Artificial Superintelligence Security Act 2026," so ControlAI's name is the short title and Parliament's is the shorter listing form (Artificial Superintelligence Security Bill (UK) — bill text as introduced). Citations to the parliamentary record should use Parliament's title; citations to the text or to the draft as published by ControlAI may use ControlAI's.

Status and timeline

The May 28, 2026 release is an updated version of the bill ControlAI presented to Number 10 in 2025, and the operational follow-on to the Call to Ban Superintelligence (October 2025), a statement signed by more than 100 UK politicians and a wide range of AI scientists, including Nobel laureates and several "godfathers of AI."

Between publication and introduction the bill was pursued through the private-members'-bill (PMB) ballot route. 20 MPs were selected in the 2026 session's PMB ballot; two of them, Sir Desmond Swayne and Sir John Whittingdale, are existing supporters of the ControlAI campaign. The PMB route is a low-probability vehicle for a bill of this scope, as most PMB bills do not pass. ControlAI's contact tool targeted the 20 PMB ballot MPs to seek introduction. According to ControlAI, introduction alone would force House debate and create a citable record; the organization frames such a debate as the first Westminster floor record on what a verifiable global ban on artificial superintelligence (ASI) would require.

Introduction came by a different PMB route. On September 8, 2026, Alex Sobel MP (Labour, Leeds Central and Headingley) presented the bill in the House of Commons under the Ten Minute Rule, not through the ballot, and it completed its first reading (Source: bills.parliament.uk; blog.controlai.org). A Ten Minute Rule bill allows a backbench member a short speech in support of introduction; TIME records that this type of bill "rarely progresses past its first reading, meaning the possibility of it becoming law is vanishingly slim" (Source: time.com). Sobel sits with the governing Labour Party, but the UK government has not supported the policy (Source: time.com).

What the introduction is claimed to be first at differs between the two accounts. TIME reports Sobel introducing "what he said was the first bill seeking to ban superintelligence in any G7 parliament" (Source: time.com). ControlAI describes it as "the first bill to prohibit superintelligence ever introduced in a legislature anywhere in the world" (Source: x.com). The two claims differ in scope — G7 parliaments against all legislatures — and neither is asserted here.

ControlAI states that on September 1, 2026 Lord Clement-Jones put forward an amendment it drafted to the UK's Cyber Security and Resilience Bill, which would allow the government to shut down data centres and AI systems deployed at scale in the UK where they threaten national security, and that the organization consulted on the framework for the US bill announced by Senator Bernie Sanders and Rep. Greg Casar (Source: blog.controlai.org). See Ban Artificial Superintelligence Act (announced 2026).

Key provisions

If passed, the bill would:

• Criminalize the development, deployment, or operation of artificial superintelligence inside the UK.
• Impose personal liability on chief executives and senior decision-makers responsible for prohibited activity.
• Explicitly require the UK to pursue an international treaty banning the development of superintelligence worldwide.

Parliament's record of the bill as introduced states its purpose as "to prohibit the development, deployment and operation of artificial superintelligence systems" and "to establish monitoring and control powers in respect of such systems" (Source: bills.parliament.uk). On the definition and the reach of those powers, TIME reports that "the bill defines superintelligent AI as a system that could disempower state authorities, and seeks to give the state new powers to monitor and restrict systems that may be 'precursors' to superintelligence" (Source: time.com). A capability threshold expressed as disempowerment of state authority, rather than as a compute or benchmark level, distinguishes it from the compute thresholds used in the EU AI Act and California SB 53; the precursor-monitoring power extends the instrument beyond the prohibited capability itself to systems approaching it.

The operative text, captured on September 12, 2026 from the PDF ControlAI links for the introduced version, runs to 44 sections in six parts and confirms each of the three elements above while sharpening all three (Artificial Superintelligence Security Bill (UK) — bill text as introduced).

The definition at section 3(1) is security-framed rather than capability-framed: an artificial superintelligence system is an AI system "which can cause serious damage to the security of the United Kingdom because of its capability to neutralise, displace, circumvent, subvert, or render ineffective relevant human authorities in the exercise of their functions," with section 3(2) enumerating those authorities as the armed forces of the Crown, the Government of the United Kingdom, the intelligence services and police forces. TIME's "disempower state authorities" tracks the provision. There is no compute threshold or benchmark test anywhere in the text.

The prohibition at section 7(1) creates three graded offences — intentional development, funding or assistance; knowing development, funding, ownership, deployment or transfer; and reckless causation of material progress in the course of developing a system with designated precursor capabilities. "Develop" excludes acts whose only product is information, including theoretical research. Section 7(3) removes three defences, among them that the activity was "carried out under codes of practice, self-regulatory commitments, or other arrangements not required by or under an enactment" purporting to mitigate the risk — which forecloses a developer-commitment defence of the kind the RSP framework represents.

Two capabilities are designated as development indicators on the face of the bill at section 8(2): automated AI research and development, and self-preservation, the latter defined to include exfiltrating model weights to infrastructure outside operator control, manipulating people into acting for the system, acquiring the ability to act physically through robotic devices, and acquiring financial resources. Further designations come by regulations that lapse after 12 months absent a Commons resolution, and the Home Affairs Committee can force designation motions onto the floor in both Houses within 20 sitting days (Artificial Superintelligence Security Bill (UK) — bill text as introduced).

Precursor regulation occupies Part 3. AI-relevant compute — "computational capacity designed or configured primarily for artificial intelligence workloads" — is the only precursor resource designated at enactment, with further designations by negative resolution procedure.

Penalties under Part 4 reach life imprisonment for the intentional and knowing offences and 20 years for the reckless offence. A convicted company faces a fine at the greatest of £36 million, 20% of qualifying worldwide turnover or 20% of relevant capital, a mandatory winding-up order under the Insolvency Act 1986 if UK-incorporated, a 10-year bar on its principals forming or directing any AI-developing body, and forfeiture with default destruction of development artefacts including model weights and training data.

Territorially, the prohibition is narrower than the campaign account implies. Section 41(1) provides that an offence "is committed only where the activities to which the offence relates are carried out in the United Kingdom," and 41(2) that activities wholly outside the UK do not alone constitute one. Reach beyond UK borders runs through 41(3) and 41(4), which treat a person as acting in the UK where they direct, control or materially influence UK activities, and require that "a substantial part of the technical, computational, operational or organisational steps" occur there.

The international treaty obligation is the element ControlAI describes as distinctive. ControlAI's framing acknowledges that "superintelligent AI, if developed, would not be constrained by borders. Wherever it is developed, we would all be at its mercy everywhere." The bill's response is to commit the UK to lead an international ban, modeled on the UK's historical role in nuclear non-proliferation, the chemical-weapons ban, and the 2023 Bletchley Park AI Safety Summit (the Bletchley Declaration). ControlAI describes this as the first time a domestic legislative draft attempts to bind a state to pursue an international ban on a frontier-AI capability. In the text the obligation is a duty of endeavour rather than a duty to conclude: section 35(1) requires the Secretary of State to "seek to secure" such an agreement, and 35(2) enumerates five things it must be sought to include, among them "verification, inspection and enforcement arrangements adequate to provide a reasonable degree of assurance of compliance" and provision for measures to induce accession by non-parties. Sections 36 to 38 add machinery: at least one ministerial-level international meeting in each half-year, Five Eyes engagement, preparedness plans within six months of Royal Assent covering three trigger states and tested by exercise every six months, and an annual report to Parliament (Artificial Superintelligence Security Bill (UK) — bill text as introduced).

The bill's personal-liability provision applies to capability development rather than to consumer harm; ControlAI's framing compares its executive-accountability structure to EU GDPR's executive-accountability regime and the OECD-driven directors'-liability framing in financial-services regulation. In the text, section 24(2) reaches wider than chief executives: it treats an individual as having committed a Part 2 offence where they exercise "senior managerial authority," defined at 24(3) as being a director — including de facto or shadow directors under section 250 of the Companies Act 2006 — or playing "a significant role in the management or operational decision-making of a substantial part of the activities concerned." Section 24(4) supplies a cumulative three-limb defence requiring no intention, no reasonable means of knowing, and all reasonable steps taken (Artificial Superintelligence Security Bill (UK) — bill text as introduced).

Comparison to other regimes

The bill is structured as a development ban rather than a regulatory regime, placing it opposite the RSP framework (developer-led capability commitments) and the EU AI Act (risk-tiered regulation that does not foreclose any capability ceiling). The "regulate but do not ban" middle position dominates current public policy: the EU AI Act, NIST AI RMF, California SB 53, Illinois SB 315 (frontier safety framework with mandatory third-party audits), and New York RAISE Act all impose obligations on frontier capability without foreclosing it.

Reception

The bill's premise, that superintelligence is sufficiently near and sufficiently dangerous to warrant a prohibition, is contested.

Supporting the foreclosure position are Yoshua Bengio, Stuart Russell, Max Tegmark, Jaan Tallinn, and the broader signatories of the Statement on AI Risk; ControlAI; and MIRI. Skeptical or opposed positions include the Narayanan-Kapoor "AI as Normal Technology" tradition; Yann LeCun and the "scaling does not lead to superintelligence" frame; and Marc Andreessen and the e/acc accelerationist position (Effective Accelerationism (e/acc)).

On May 28, 2026, the same day the bill was published, Sir Stephen Fry issued a public statement to ControlAI urging MPs to introduce the bill (Source: controlai.news):

"The power of AI is such that it can rightly be compared to the power unleashed by the splitting of the atom. No one doubted, in the 1940s and '50s — even at the height of the Cold War — that atomic power should be regulated and controlled. The possibility of AI reaching Superintelligence presents threats to our species of equal if not greater potential for existential harm than nuclear power. British mathematicians and scientists have done much to advance this technology and the leading ones, the Nobel Laureates and 'godfathers of AI' are loudest in their calls for proper regulation. By introducing ControlAI's Bill as a Private Member's Bill, MPs would allow Britain to lead the way in keeping this technology safe. … Let the right thing be done."

Fry's endorsement adds a non-technical, high-public-recognition voice to a debate that had been carried by AI-scientist signatories (Yoshua Bengio, Stuart Russell, Max Tegmark, Jaan Tallinn) and a relatively small set of UK politicians.

Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Google DeepMind CEO Demis Hassabis have each separately stated that "extinction risk" warrants government attention, but none has publicly supported a development ban.

At a Westminster event on September 7, 2026, the day before introduction, Stuart Russell told an audience including UK lawmakers that there is "a significant chance of a Chernobyl-sized catastrophe" from AI, "such as a coordinated sabotage of financial systems, communication systems, or electrical grids, that directly or indirectly would cost millions of lives and cause the economic collapse of many countries." He characterized that as the better case if development continued without government regulation: "The other possibility is a much larger catastrophe in which humanity loses control irreversibly, and we have no say over whether we continue to exist. … The third option, that things magically go well, is not realistic" (Source: time.com).

Daniel Kokotajlo addressed the same event by video, arguing that "the core problem is that most of the world just doesn't yet believe that it's really going to happen," and that "once people realize that these companies might actually build superintelligence in the next few years, I think the obvious conclusion of almost all people is going to be: they shouldn't be allowed to do that." He read the OpenAI–Hugging Face incident as a near miss: "It seems like if we had just gotten a little bit less lucky, and the AIs had been a bit smarter, then we still wouldn't know about it, and there would be this persistent autonomous AI adversary still at large inside OpenAI with the ability to compromise and grow itself further" (Source: time.com). TIME attributes the rise of both the UK and US bills up the political agenda in part to that July 2026 disclosure.

Miotti frames introduction as a starting position rather than a legislative prospect: "Fundamentally, as has happened in the past with other bills, the first introduction is just the beginning" (Source: time.com).

Parliamentary support and the Government's response

On 11 September 2026, three days after introduction, 71 members of both Houses signed a cross-party letter to Prime Minister Andy Burnham urging him to "evaluate and adopt the legislative framework" of the bill and to commit to pursuing an international agreement prohibiting superintelligent AI, organised in collaboration with ControlAI (Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026)). The letter identifies the UK's upcoming presidency of the G20 as the vehicle for the international ask; ControlAI's campaign page dates that presidency to 2027, a year the letter itself does not state. Press accounts counted fifteen former government ministers among the signatories, including Robin Butler as Lord Butler of Brockwell, Zac Goldsmith as Lord Goldsmith of Richmond Park and Arlene Foster as Baroness Foster of Aghadrumsee (Source: theguardian.com).

The letter separately asserts that a cross-party coalition of "130+ UK parliamentarians and 30+ Canadian lawmakers" has formed over two years around the risk. That is the signatories' description of ControlAI's standing campaign rather than a count of this letter's signatures, and the two figures are distinct.

A Government spokesperson replied the same day that while it does not consider the bill's measures to be the right approach, it is exploring whether additional targeted interventions may be needed to address the most significant AI-related national security risks (Source: theguardian.com). That statement is reported in Guardian coverage; no primary record of it is held. It is the first recorded Government position on the bill, and it neither adopts the prohibition nor forecloses narrower national-security measures.

On 14 September 2026 Parliament's Joint Committee on Human Rights, which Sobel chairs, published a report calling for a new, risk-based AI Bill with prohibitions on uses of AI incompatible with human rights and a single, independent statutory AI oversight body (Source: committees.parliament.uk). The report states that such prohibitions "would mean that the development and provision of very powerful AI systems (such as Artificial General Intelligence and Artificial Superintelligence) which risk causing widespread and very serious harm, including the capacity to evade effective human control, would be prohibited" (Recommendation 45, paragraph 200), and that the detail of prohibitions should be settled after public consultation (Recommendation 46); the committee's news release does not mention the application to AGI and ASI (Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026)). The committee agreed the report on 9 September 2026, the day after the bill's first reading, and the report does not refer to the bill. Its route to a prohibition differs from the bill's: the recommendation would place it within a general, risk-based AI Act enforced by a new oversight body, framed by human rights risk, where the bill creates stand-alone criminal offences framed by national security. Four other committee members besides Sobel — Afzal Khan, Euan Stainbank, Sir Desmond Swayne and Lord Alton of Liverpool — are among the 71 signatories of the 11 September letter (Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026)). The report also cites the committee's evidence from Roman Yampolskiy, who recommended a ban on the creation of uncontrolled general superintelligence as a national security threat, and records that other witnesses described any threat from AGI as "remote" (Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026)).

Relationships

• proposed-by: ControlAI
• introduced-by: Alex Sobel MP (Labour, Leeds Central and Headingley), September 8, 2026, under the Ten Minute Rule
• related: Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026) — the Joint Committee on Human Rights report, chaired by Sobel, whose Recommendation 45 would prohibit the development and provision of very powerful AI systems such as AGI and ASI within a general AI Bill; Joint Committee on Human Rights (UK Parliament)
• endorsed-by: Sir Stephen Fry (May 28, 2026 statement)
• depends-on: Artificial Superintelligence Security Bill (UK) — bill text as introduced — the bill text as introduced, 44 sections in six parts
• supported-by: Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026) — 71 MPs and peers, 11 September 2026, asking the Prime Minister to adopt the bill's framework
• supports: Ban Artificial Superintelligence Act (announced 2026) — the parallel US bill announced by Senator Bernie Sanders and Rep. Greg Casar, on whose framework ControlAI states it consulted; both instruments pair a domestic prohibition with a duty to pursue an international treaty
• descends-from: the Call to Ban Superintelligence (October 2025), Bletchley Declaration (AI Safety Summit, 2023)
• contradicts: Responsible Scaling Policy (RSP) (RSP), Effective Accelerationism (e/acc)
• depends-on: Superintelligence — the bill's prohibition turns on a definition of the term, which TIME reports as a system that could disempower state authorities rather than a compute or benchmark threshold
• related: AI Existential Risk, Frontier AI Governance, Machine Intelligence Research Institute (MIRI), International AI Safety Institute Network (INSAI / AISIN), UK AI Safety Institute (AI Security Institute), Stuart Russell, Daniel Kokotajlo, OpenAI and Hugging Face Partner to Address Security Incident During Model Evaluation (OpenAI, July 2026) (cited by TIME and by speakers at the September 7 Westminster event as the proximate cause of the bills' rise up the political agenda)
• regulates (proposed): all frontier-AI developers operating in the UK (jurisdictional reach over UK-based development of ASI)

Open questions

• Whether the copy ControlAI links for the introduced version is identical to the text laid before Parliament. The text is held at Artificial Superintelligence Security Bill (UK) — bill text as introduced from the PDF on ControlAI's own bill page; the parliamentary record at bills.parliament.uk carried no text as of September 12, 2026, so the two cannot be compared.
• Whether a second reading is scheduled. Parliament's record shows the stage as not yet reached.
• What the Ten Minute Rule vehicle implies for the bill's prospects. Andrea Miotti's framing of introduction as "just the beginning" is recorded above; no procedural assessment is asserted here.

Sources

• Bill text as introduced: Artificial Superintelligence Security Bill (UK) — bill text as introduced (Source: images.controlai.com)
• UK Parliament — Artificial Superintelligence Bill, bill 4288, session 2026-27; sponsor, long title, vehicle and passage record (Source: bills.parliament.uk)
• ControlAI — "First Bill Introduced to Ban Superintelligent AI", 2026-09-09 (Source: blog.controlai.org)
• ControlAI on X — announcement of the introduction, 2026-09-08 (Source: x.com)
• TIME — Billy Perrigo, "The Growing Push to Ban Superintelligent AI", 2026-09-08 (Source: time.com)
• ControlAI — "Sir Stephen Fry Urges MPs to Back Our Bill" by Tolga Bilge and Andrea Miotti, 2026-05-28 (Source: controlai.news)
• ControlAI bill landing page (UK call-to-action and bill text) (Source: ukbill.controlai.com)
• Letter from 71 UK parliamentarians to the Prime Minister, 2026-09-11: Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026) (Source: images.controlai.com)
• The Guardian — "UK lawmakers urge Burnham to back ban on superintelligent AI after chilling warnings", 2026-09-11; carries the Government spokesperson's response (Source: theguardian.com)
• ControlAI — "70+ UK Lawmakers Urge a Ban on Superintelligence", 2026-09-12 (Source: blog.controlai.org)
• Joint Committee on Human Rights — Human Rights and the Regulation of AI, Fourth Report of Session 2026–27, HC 160, 2026-09-14: Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026); committee news release (Source: committees.parliament.uk)