The Artificial Superintelligence Security Bill is a United Kingdom private member's bill drafted by ControlAI and introduced in the House of Commons by Alex Sobel MP on 8 September 2026 under the Ten Minute Rule, with cross-party co-sponsorship. It runs to 44 sections in six parts and would prohibit the development, deployment and operation of artificial superintelligence systems in the United Kingdom, establish a monitoring and restriction regime for precursors to such systems, and oblige the Secretary of State to seek an international prohibition agreement. Section 44 gives the short title as the Artificial Superintelligence Security Act 2026; section 42 commences the Act on Royal Assent; section 43 extends it to the whole of the United Kingdom. For the bill's political history, sponsorship and reception, see UK Artificial Superintelligence Bill (ControlAI draft, May 28, 2026; introduced September 8, 2026).
Purpose and definitions
Section 1 states three purposes: effective prevention of ASI development within the United Kingdom, effective monitoring and restriction of ASI precursors within the United Kingdom, and "demonstrable progress towards an international agreement" for prohibition and precursor control.
The operative definition is security-framed rather than capability-framed. Section 3(1) defines an "artificial superintelligence system" as 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." Section 3(2) enumerates those authorities as the armed forces of the Crown, the Government of the United Kingdom, the intelligence services and police forces. The bill sets no compute threshold, benchmark score or general-intelligence test.
Section 2 takes the OECD-style definition of an AI system — a machine-based system that "for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments."
Three further terms carry weight. An "artificial superintelligence development indicator" (section 4) is an AI capability that both substantially increases the likelihood or feasibility of ASI development, or is integral to a means by which an ASI system could threaten UK security, and "would of itself fundamentally undermine the effective prevention" of ASI development. An "ASI precursor" (section 5) is "any resource, AI capability, input, or method that may reasonably be expected to be used in, or to materially assist," ASI development. Section 6 defines capability possession to include feasible modification, and defines independent exercise of a capability as one where human involvement is "nominal or perfunctory," limited to providing resources or execution at the system's direction, or "otherwise insufficient to constitute meaningful human direction or control over the material steps."
Prohibition (Part 2)
Section 7(1) creates three offences. Under 7(1)(a), developing an AI system, funding AI development, or providing advice or assistance to another's AI development, intending that an ASI system or designated development indicator be developed. Under 7(1)(b), knowingly developing a system known to constitute or be "near certain to result in" an ASI system or designated indicator; funding a person known to be developing or attempting to develop one; or owning, deploying, transferring or otherwise providing one. Under 7(1)(c), recklessly causing material progress towards an ASI system or designated indicator in the course of developing a system with designated precursor capabilities.
"Develop" is defined at 7(2)(a) as performing, directing or participating in "any concrete act of creating, constructing, training or modifying an AI system," with two carve-outs: acts whose only product is information, including theoretical research, and non-substantial modifications made solely to determine a system's capabilities. "Fund" at 7(2)(b) covers capital or resources provided "by any means, directly or indirectly."
Section 7(3) removes three defences to the intentional and knowing offences: that the activities were intended to produce benefits of any kind; that they were carried out under "codes of practice, self-regulatory commitments, or other arrangements not required by or under an enactment" purporting to mitigate ASI risk; or that they were "an adjunct to research for other purposes." Section 7(4) supplies a compliance safe harbour against the recklessness offence for a person who complied with the section 15(1) requirements, or who took reasonable precautions before that regime existed — disapplied by 7(5) where the person was aware of a risk that it was unreasonable to take.
Section 8(2) designates two indicators on the face of the bill: automated AI research and development, being the capability to carry out AI research and development independently and to a substantial extent, including generating or refining research hypotheses, architectures, training methods or experimental designs, designing or training or improving AI systems or components, and designing or producing hardware or compute infrastructure primarily intended for AI systems; and self-preservation, being the capability to ensure continued independent operation despite attempts to shut the system down, including by exfiltrating model weights and software to infrastructure outside operator control, compromising or coercing or manipulating people to act for the system, acquiring the capability to act physically through robotic or autonomous devices, and acquiring financial resources to purchase hardware, services or labour.
Further designations come by regulations under section 8(3), subject to constraints: written notice so far as reasonably practicable to persons developing or operating a system likely to exhibit the indicator (8(4)); a seven-day delay before the designation bites for section 7 purposes (8(5)); lapse after 12 months unless approved by resolution of the House of Commons (8(6)); and, on lapse, a bar on bringing or continuing proceedings or sustaining convictions that depend on the lapsed designation (8(7)(a)). Sections 8(8) and 8(9) give the Home Affairs Committee a forcing role: a committee resolution recommending additional designations obliges the Secretary of State to table and make time for endorsing motions in both Houses within 20 sitting days, and an endorsing resolution in either House obliges designation within 30 days.
Section 9 makes Part 2 offences continuing offences, committed afresh on each day the activity continues. Section 10 binds the Crown.
Precursor monitoring and restriction (Part 3)
Section 11(2)(a) designates a single precursor resource at enactment: "AI-relevant compute, that is computational capacity designed or configured primarily for artificial intelligence workloads." Additional resources may be designated by regulations under the negative resolution procedure (sections 11(3), 11(4)). Sections 12 through 15 govern designation of precursor capabilities, monitoring, and restrictions on designated resources and capabilities respectively. Section 16 gives the Comptroller and Auditor General a reporting function; section 17 provides for parliamentary imposition of designations and regulations; section 18 extends monitoring to government AI projects. Section 19 treats contraventions as continuing; section 20 binds the Crown.
Enforcement and penalties (Part 4)
| Conduct | Penalty |
|---|---|
| Section 7(1)(a) or 7(1)(b) offence, individual | Imprisonment for life on conviction on indictment (s 21(1)) |
| Section 7(1)(c) offence, individual | Imprisonment up to 20 years (s 21(2)) |
| Section 7(1) offence, non-individual | Fine at the greatest of £36 million, 20% of qualifying worldwide turnover for the preceding financial year, or 20% of relevant capital (s 21(3)) |
| Continuing section 7(1) offence, non-individual | Further daily fine at the greatest of £1 million, 5% of turnover, or 5% of relevant capital (s 21(4)) |
| Contravention of section 14 or 15 | Civil penalty at the greatest of £100,000, 2% of turnover, or 2% of relevant capital (s 22(1)) |
| Continuing contravention of section 14 or 15 | Further daily penalty at the greatest of £5,000, 0.5% of turnover, or 0.5% of relevant capital (s 22(2)) |
"Relevant capital" is defined within sections 21 and 22 as the higher of gross assets or the total capital and funding raised for or applicable to the development.
Three further consequences attach to conviction of a non-individual. Section 21(5) requires the court, for a body incorporated or registered in the United Kingdom, to order winding-up under the Insolvency Act 1986 as if a public-interest petition had been presented under section 124A, with power to appoint a liquidator. Section 21(6) bars a convicted person for ten years from forming, holding a controlling interest in, or acting as a director — including de facto or shadow director within section 250 of the Companies Act 2006 — of any body engaged in developing, training or deploying an AI system, enforced through the Company Directors Disqualification Act 1986 (s 21(7)). Sections 21(8) to 21(10) require forfeiture to the Government of any "development or deployment artefact" relevant to the offence — defined to include model weights, parameters and trained or partially trained models, training and evaluation datasets, source code and tools, hardware and computing infrastructure, technical documentation and research records, and any copy, derivative or backup — and require the court to order destruction unless retention is necessary for evidential or investigative purposes.
Section 24 extends liability to individuals. Subsection (1) treats anyone wholly or partly responsible for, or participating in, another's offence as having committed it. Subsection (2) treats an individual as having committed a Part 2 offence where they exercise "senior managerial authority" over the constituting activities, defined at 24(3) as being a director (including de facto or shadow directors) or playing "a significant role in the management or operational decision-making of a substantial part of the activities concerned." Subsection (4) provides a cumulative three-limb defence: no intention to undertake or participate in the activities, no reasonable means of knowing, and all reasonable steps taken to avoid it.
Sections 25 and 26 each require the Secretary of State to make regulations within six months of Royal Assent — a self-reporting regime designed so that fear of prosecution does not deter reporting of "accidental or unintended occurrences, results, or activities," with reduced liability for information that materially assists investigation of another's offence when provided before becoming a subject of investigation (s 25(4)); and a whistle-blowing regime protecting individuals who report offences or contraventions. Sections 27 to 34 provide the enforcement toolkit: challenge inspections, suspension notices, preservation notices, a duty to provide assistance, an emergency direction on materialisation of risk, continuing offences, penalties for failure to comply with enforcement measures, and delegation of enforcement functions.
International coordination (Part 5)
Section 35(1) obliges the Secretary of State to "seek to secure" an international agreement whose principal purpose is worldwide prohibition of ASI development, deployment and operation together with precursor monitoring and restriction. Section 35(2) requires that the agreement be sought to include prohibition by all parties, precursor monitoring and restriction, "verification, inspection and enforcement arrangements adequate to provide a reasonable degree of assurance of compliance," accession provision for non-signatories, and provision for measures to induce accession by non-parties. It is a duty of endeavour rather than a duty to conclude.
Section 36 requires an international coordination programme including at least one ministerial-level international meeting in each half of a calendar year, additional officials-level and technical meetings at discretion, and engagement with the Five Eyes governments and such other states and international organisations as the Secretary of State considers appropriate. Section 37 requires preparedness plans within six months of Royal Assent for three trigger states — precursor capabilities approaching a level posing material risk to UK security, activity likely to result in an ASI system, or an ASI system having been or being developed, deployed or operated, whether inside or outside the UK. Those plans must cover expedited suspension and preservation notices, coordination of the armed forces, intelligence services, police and other public authorities, continuity of government functions and critical national infrastructure, information-sharing with the Five Eyes, and public communication; they must be reviewed each reporting period and after any materially affecting development, tested by exercise at least every six months, summarised to Parliament subject to national-security and international-relations limits, and supplied in full to the Intelligence and Security Committee. Section 38 requires an annual international report laid before Parliament.
Territorial application
Section 41(1) limits offences to activities carried out in the United Kingdom, and 41(2) provides that activities carried out wholly outside the United Kingdom do not alone constitute an offence. Reach beyond UK borders is indirect: 41(3) treats a person as engaging in UK activities where they direct, control or materially influence the carrying out of those activities in the UK, or cause them to be carried out there; 41(4) treats activities as carried out in the UK where "a substantial part of the technical, computational, operational or organisational steps" occurs there; and 41(5) lists factors the court may weigh in assessing direction or control — governance, approval or funding decisions, project authorisation or termination powers, allocation of personnel, compute or infrastructure, and setting or approval of technical objectives.
Provenance
Captured on 2026-09-12 from the PDF linked from ControlAI's bill page, complete at 22 pages and all 44 sections. ControlAI's page states the bill was introduced in the House of Commons by Alex Sobel MP on 8 September 2026 with cross-party co-sponsorship, and describes it as "the first piece of legislation in any legislature designed to prohibit the development of superintelligent AI." Ten divergences between the text and the secondary accounts the vault previously relied on are recorded in the raw file's ## Source-fidelity notes; the three most consequential are that the bill's own short title matches ControlAI's name for it rather than contradicting it, that the prohibition is territorially limited with only indirect extraterritorial reach, and that individual liability attaches to any senior manager rather than only to chief executives and carries a three-limb defence.
Relationships
- supports: UK Artificial Superintelligence Bill (ControlAI draft, May 28, 2026; introduced September 8, 2026) — the primary text behind the page built from ControlAI's account and press coverage
- related: ControlAI — the drafter
- related: Ban Artificial Superintelligence Act (announced 2026) — the US framework ControlAI states it consulted on, for comparison of prohibition designs
- related: Superintelligence — the subject the bill defines by reference to UK security rather than capability
- related: Recursive Self-Improvement (RSI) — automated AI research and development, designated on the face of the bill at section 8(2)(a)
- related: Compute Governance — AI-relevant compute as the sole precursor resource designated at enactment
- related: AI Autonomy Risk — self-preservation, designated at section 8(2)(b), including weight exfiltration and resource acquisition
Sources
- Raw source:
Raw Sources/Artificial Superintelligence Security Bill (UK, ControlAI draft as introduced 2026-09-08).md - Bill text PDF: images.controlai.com
- ControlAI bill page: controlai.org