ControlAI is a UK-based AI-policy advocacy organization that describes its mission as "working to keep humanity in control" of artificial intelligence development. Its primary focus is preventing the development of artificial superintelligence, and it frames its work around AI existential risk. Rather than publishing policy papers, the organization drafts statutory text and runs constituent-contact campaigns, with outputs that include the UK Artificial Superintelligence Security Bill and cross-party legislator statements in the UK and Canada.
Leadership
- Andrea Miotti — founder, CEO, and US Director of advocacy. Miotti has appeared on Al Jazeera's Upfront with Redi Tlhabi (mid-2026).
- Tolga Bilge — co-author of the ControlAI Substack newsletter; runs the bill-introduction campaign.
- Connor Leahy — US Director (per the May 28, 2026 ControlAI dispatch).
- Sophie Toura — represented ControlAI at the International Scientific Exchange for AI Safety in Singapore in May 2026, alongside Yoshua Bengio, Dawn Song, Stuart Russell, Jaan Tallinn, and Max Tegmark.
Activities
Artificial Superintelligence Security Bill (UK)
ControlAI authored the UK Artificial Superintelligence Security Bill, a legislative draft that would criminalize the development of artificial superintelligence. The version published on May 28, 2026 is an update of the bill ControlAI presented to Number 10 in 2025. Sir Stephen Fry endorsed the bill the same day; ControlAI characterizes the endorsement as evidence of uptake of the existential-risk framing by a mainstream cultural figure, beyond the prior pattern of endorsement by AI scientists.
ControlAI is targeting the 20 MPs selected in the 2026 session's private-members'-bill (PMB) ballot — including supporters Sir Desmond Swayne and Sir John Whittingdale — to introduce the bill to Parliament. The bill was introduced in the House of Commons by Alex Sobel MP on September 8, 2026 under the Ten Minute Rule, with cross-party co-sponsorship, and ControlAI describes it as "the first piece of legislation in any legislature designed to prohibit the development of superintelligent AI" (Source: controlai.org).
Three days after introduction, ControlAI organised a cross-party letter to Prime Minister Andy Burnham signed by 71 MPs and peers, urging him to adopt the bill's legislative framework and to pursue an international prohibition agreement through the UK's upcoming G20 presidency; the organisation hosts the letter and published an account of it on 12 September 2026 (Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026); Source: blog.controlai.org). The letter states that the coalition ControlAI has assembled over two years now numbers "130+ UK parliamentarians and 30+ Canadian lawmakers" — an organisational claim, and the most recent figure the campaign has published, against "over 120 UK lawmakers" in July 2026. A Government spokesperson replied that it does not consider the bill's measures to be the right approach while it explores whether additional targeted interventions are needed for the most significant AI-related national security risks (Source: theguardian.com).
The drafted text runs to 44 sections in six parts and is held at Artificial Superintelligence Security Bill (UK) — bill text as introduced. It is the organization's most detailed published policy instrument, and its drafting choices are attributable to ControlAI rather than to Parliament: a definition of artificial superintelligence keyed to damage to UK security through the neutralisation of named state authorities rather than to any compute or benchmark threshold; automated AI research and development and self-preservation designated as development indicators on the face of the bill; AI-relevant compute as the sole designated precursor resource; criminal penalties reaching life imprisonment with senior-manager liability, mandatory winding-up of a convicted UK company and forfeiture with default destruction of model weights; and a duty on the Secretary of State to "seek to secure" an international prohibition agreement with specified contents. The text's own short title, at section 44, is the Artificial Superintelligence Security Act 2026, which makes ControlAI's name for the instrument the bill's own rather than a campaign label (Artificial Superintelligence Security Bill (UK) — bill text as introduced).
ControlAI's bill page also carries endorsements from Lord Browne of Ladyton, the former Defence Secretary, who argues that "[o]nly binding law can stop leading AI companies from racing to build superintelligence"; Beatrice Fihn, executive director of Lex International and former executive director of ICAN; Stuart Russell; and Daniel Kokotajlo, who argues that even a successful attempt would create "the most intense concentration of power in human history" (Source: controlai.org).
Cross-party legislator statements
ControlAI's Call to Ban Superintelligence, which states "We call for a prohibition on the development of superintelligence", was signed by more than 100 UK politicians together with AI scientists, Nobel laureates, and figures sometimes described as the "godfathers of AI" (Yoshua Bengio, Geoffrey Hinton).
The coalition has grown across two jurisdictions and is now counted by ControlAI at "130+ UK parliamentarians and 30+ Canadian lawmakers," per the September 2026 letter it organised, against "over 120 UK lawmakers" stated in July 2026 (Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026)).
The Canadian call
On June 4, 2026, ControlAI announced that more than 30 Canadian MPs and senators — spanning Liberal, Conservative, Bloc, and Independent ranks — are calling for Canada to negotiate an international "trust but verify" agreement prohibiting the development of superintelligence. The Canadian statement extends the UK template (the Call to Ban Superintelligence and the ASI Security Bill) into a second G7 jurisdiction and reframes the ask around international verification rather than purely domestic prohibition. It was announced the same day as Anthropic's pause-verification-infrastructure argument, with the two converging on the verification problem while differing on prohibition. See Frontier AI Governance. (Source: cbc.ca)
Parliamentary evidence
Andrea Miotti submitted written evidence jointly with Steven Adler, a former OpenAI researcher who co-led the development of dangerous-capability evaluations, to the UK Parliament's Joint Committee on Human Rights inquiry into human rights and the regulation of AI (reference RAI0031). The committee's September 2026 report cites the submission among the witnesses who raised concerns about long-term risks from artificial general intelligence, and as the technical evidence behind its statement that "many in the AI industry consider these voluntary commitments" — developers' published safety frameworks and "intolerable" risk thresholds — "inadequate, unreliable and ineffective, particularly given the strong financial incentives driving firms to avoid oversight." The report recommends a statutory AI Security Institute with mandatory pre-release submission of powerful models, and a general AI Bill whose prohibitions would extend to the development and provision of very powerful systems such as AGI and ASI (Recommendations 45, 57–59) (Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026)). The committee is chaired by Alex Sobel MP, who introduced ControlAI's bill.
Published argument
On July 4, 2026, ControlAI published an essay by Lord Fairfax of Cameron calling on the next UK government to prohibit superintelligence development domestically and to pursue an international prohibition agreement. The essay cites MI5's late-2025 warning about autonomous AI systems evading oversight and notes that over 120 UK lawmakers back ControlAI's campaign recognizing superintelligence as a national and global security threat (Source: controlai.news).
Frontier framework stopping conditions
On August 14, 2026, Tolga Bilge and Andrea Miotti published an argument that neither leading laboratory's published framework obliges it to halt development at high cyber capability. On their reading, OpenAI's Preparedness Framework says the company should halt at "Critical" cyber capability but is not legally binding, while the Frontier Governance Framework OpenAI must publish under California SB 53 — which carries civil penalties of up to $1 million per violation — permits continued development at the equivalent Tier 3 where residual risk is judged acceptable. They further note that Anthropic's cyber trigger was narrowed to an "ongoing assessment" in RSP 2.0 and removed in RSP 3.0, published February 24, 2026 — the same day Claude Mythos Preview was first made available internally — and that the word "cyber" does not appear in RSP 3.0 (Source: controlai.news).
The argument is a position advanced by an advocacy organization campaigning for prohibition, not an independent compliance finding, and neither company has responded to it on the record. It is one of the few published readings that treats the binding force of a framework's stopping condition, rather than its threshold definitions, as the operative question. See AI Safety Cases and Frameworks and Autonomous cyber-agents.
Constituent-contact tooling
ControlAI runs contact tools that let UK and US supporters write to their elected representatives in roughly a minute. The UK-specific bill-introduction tool targets the 20 PMB ballot MPs. The Microcommit project sends supporters a small number of weekly tasks. This constituent-contact infrastructure is comparable to post-SB 1047 organizing tools built by Encode Justice and others.
Media presence
The ControlAI Substack newsletter had over 148,000 subscribers as of May 2026, placing it among the larger AI-policy newsletters and comparable in scale to Casey Newton's Platformer. ControlAI's leaders appear on broadcast and online media: Connor Leahy on Al Jazeera's Upfront (May 2026); Andrea Miotti on the Professor Dave Explains YouTube channel (4.31M subscribers) in mid-2026; and Sophie Toura at the IMDA-convened International Scientific Exchange for AI Safety in Singapore.
Position in the AI-policy landscape
ControlAI is single-issue, focused on prohibiting superintelligence development, in contrast to broader AI-safety policy organizations such as Center for Strategic and International Studies (CSIS) and CAIS. The organization describes this narrow scope as a deliberate choice, comparable in form to single-issue advocacy organizations on guns, abortion, or climate. Its method is legislative: it drafts statutory text and runs constituent-contact campaigns rather than publishing white papers.
ControlAI is a consistent advocate of the "ban, not regulate" position in current AI policy debates. Frameworks and bills on the regulate-but-do-not-ban side include RSP, the EU AI Act, Illinois SB 315 (frontier safety framework with mandatory third-party audits), California SB 53, and New York RAISE Act.
ControlAI's foundational claim — that superintelligence is sufficiently near and sufficiently dangerous to warrant a development prohibition — is contested by other tracked positions, including Narayanan and Kapoor, Yann LeCun, and Effective Accelerationism (e/acc). ControlAI's stated tactical theory is that even partial legislative engagement — debate, floor records, civil-society organizing — builds institutional capacity for a binding regime later.
Relationships
- authored: UK Artificial Superintelligence Bill (ControlAI draft, May 28, 2026; introduced September 8, 2026) (May 28, 2026 draft; introduced September 8, 2026)
- authored: Artificial Superintelligence Security Bill (UK) — bill text as introduced — the drafted text, 44 sections in six parts
- organised: Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026) — the 71-signatory letter to the Prime Minister, 11 September 2026
- descends-from: Bletchley Declaration (AI Safety Summit, 2023) (organizational influence), the broader AI Existential Risk community
- aligned-with: Yoshua Bengio, Stuart Russell, Max Tegmark, Jaan Tallinn, Machine Intelligence Research Institute (MIRI)
- contradicts: Responsible Scaling Policy (RSP) (the developer-led capability-commitment approach), Effective Accelerationism (e/acc)
- related: AI Existential Risk, Superintelligence, Frontier AI Governance, Sir Stephen Fry (May 28 endorser)
Sources
- ControlAI — UK Artificial Superintelligence Security Bill page, with the bill-text PDF and endorsement block (Source: controlai.org)
- ControlAI Substack — "Sir Stephen Fry Urges MPs to Back Our Bill" by Tolga Bilge and Andrea Miotti, 2026-05-28 (Source: controlai.news)
- ControlAI UK bill landing page (Source: ukbill.controlai.com)
- ControlAI — "70+ UK Lawmakers Urge a Ban on Superintelligence", 2026-09-12 (Source: blog.controlai.org); letter text at Letter from 71 UK parliamentarians to Prime Minister Andy Burnham on superintelligent AI (11 September 2026)
- The Guardian — "UK lawmakers urge Burnham to back ban on superintelligent AI after chilling warnings", 2026-09-11 (Source: theguardian.com)