Connects public persuasion to proposed AI prohibitions
ControlAI
ControlAI campaigns to prevent the development of artificial superintelligence. It combines legislative proposals, briefings for politicians, creator partnerships and public mobilization, with organizations in the UK and United States. Its work connects arguments about AI catastrophe to proposals for enforceable limits on who may build advanced systems.
ControlAI seeks to make the development of artificial superintelligence a prohibited activity. Its programme includes A Narrow Path and more specific legislative proposals. Published plans describe recruiting political allies through briefings, introductions, public statements and sustained contact. Andrea Miotti leads the organization; Connor Leahy moved into its US leadership after his work at Conjecture. The campaign draws on people with technical AI backgrounds as well as political and communications experience.
The persuasion work is now unusually visible. On 8 September 2026, Miotti and creator-outreach lead Max Winga identified ControlAI as the organization behind the approach Sabine Hossenfelder had described. Hossenfelder and ControlAI disagree about the editorial conditions attached to that offer. ControlAI also has an acknowledged partnership with Rational Animations. These relationships place its policy arguments in front of audiences well beyond parliamentary offices or specialist AI forums.
Its financial history includes two donations reported by Jaan Tallinn to its UK company: £500,000 in November 2023 and £2 million in April 2024. The company, formerly Secure Future Research, changed its registered name to CONTROLAI on 10 September 2026. Miotti and Eva Behrens have recorded voting and director-appointment rights there.
The current political reach crosses borders. ControlAI says it consulted on the Sanders–Casar framework announced in September, while Miotti distinguishes that framework’s broad pause from ControlAI’s narrower preference. PauseAI describes its own mobilization as complementary to ControlAI’s policy work. Together these records show how funding, communication, organizing and legislative access can support a sustained restriction campaign.
22-page download defines ASI through specified UK security harm and proposes compute identity records, seizure/destruction and expert oversight. Separate from emergency-shutdown amendment; not enacted.
Published Direct Institutional Plan and impact report describe introductions, dinners, briefings, signatures and public statements; claimed conversion effects not independently audited.
used its plan as a reference in the draft grant design
Draft read 7 Sep 2026
CASIR says its grant committee would initially reflect ControlAI’s institutional plan. This is CASIR’s proposed design, not verified ControlAI direction of an operating fund.
Employer posting describes partner recruitment, briefings and script work focused on extinction risk. £60k–£80k salary is advertised employee pay, not creator fees.
promotes ASI prohibition as red-lines implementation
25 September2025
ControlAI explicitly acknowledges unspecified red lines in the general call, then promotes its own prohibition and A Narrow Path. Does not bind other signatories.
ControlAI's public statement lists supporter. Does not establish endorsement of every subsequent draft/tactic, compensation, or Fabian institutional direction.
Specific output0bnxF9YfyFI names ControlAI and links its political-mobilization tools. Cash amount, script authority and actual contracting entity unresolved.
A completed ControlAI collaboration is acknowledged on both sides
The inspected SciShow video of 31 October 2025 likewise names ControlAI and its lawmaker tool in its description. The Hey AI video linked by Frame does not name CAIS in the inspected description, but Frame's claim of a CAIS contract does not establish this particular video was the contract output. Neither an absent description reference nor absent paid-promotion metadata proves a disclosure violation. Those narrower checks are preserved as boundary findings, not accusations.
Source: https://controlai.org/blog/prevent-asi-plan Title: Preventing extinction from ASI on a $50M yearly budget. By Andrea Miotti, Alex Amadori and Gabriel Alfour; April 20, 2026. Opened/read.
ControlAI reports nine paid and nine unpaid videos, around 1,000 creators contacted and more than 60 briefed. Those are its own campaign figures. The post documents its stated sponsorship process, including contracts, review, citations and disclosure; it does not demonstrate Paul's participation. See adjacent-controlai-creator-outreach.md for the competing accounts and source limits.
Separately, ControlAI's 8 September post, by Andrea Miotti and Max Winga , says it was the organization negotiating with Hossenfelder and disputes her account. The supplied Sabine transcript itself left her offeror unnamed while separately naming ControlAI as sponsoring other creators. The correct update is: ControlAI subsequently self-identified as the organization involved. Its post does not publish the underlying emails or contracts, and Paul is not named.
Lords debate attributes the emergency AI-shutdown proposal to ControlAI. Shared legislative work does not establish funding from 5Rights or its donors.
April 20 plan reports 14 collaborations, including Hank Green and Rational Animations. Individual paid/free classification, amounts and contractual rights not disclosed.
Employer posting describes partner recruitment, briefings and script work focused on extinction risk. £60k–£80k salary is advertised employee pay, not creator fees.
Published Direct Institutional Plan and impact report describe introductions, dinners, briefings, signatures and public statements; claimed conversion effects not independently audited.