AI companies are less regulated than sandwich shops

AI companies are less regulated than sandwich shops is the framing through which PauseAI names the central absurdity of the current US and UK regulatory posture toward frontier-AI labs: that a corner restaurant must demonstrate compliance with food-safety, kitchen-inspection, zoning, and licensing regimes before serving its first customer, while a company training a model on the order of magnitude that PauseAI and its peers regard as potentially civilisation-altering needs to clear no comparable safety bar before public release. The phrase entered mainstream press as a headline artefact through PauseAI organising director Ella Hughes at the 30 June 2025 picket and mock-trial outside Google DeepMind's London headquarters, was carried the same day as the verbatim Yahoo News / Business Insider headline ("Protesters accuse Google of breaking its promises on AI safety: 'AI companies are less regulated than sandwich shops'"), and has since travelled into Future of Life Institute communications, AFL-CIO labour messaging, and the wider AI-safety / pause-movement working vocabulary.

Origin

A sandwich-versus-AI regulatory analogy had been circulating in adjacent AI-safety advocacy for several months before the 30 June 2025 DeepMind protest. Yoshua Bengio used a sandwich variant ("a sandwich has more regulation than AI") in his April 2025 TED2025 talk, and Max Tegmark has used cognate formulations across 2025 — by November his settled debate-opening line ran "right now there are more regulations on sandwiches than superintelligence in the US". The construction is not a single-author coinage; the analogy was building up in the AI-safety field through 2024 and 2025 as the gap between food-safety regulation (highly mature, multi-tier, public-facing) and AI regulation (sparse, voluntary, contested) became a working talking point.

What Hughes did on 30 June 2025 was crystallise the analogy into a single sentence built for the protest line and the news cycle: "Right now, AI companies are less regulated than sandwich shops." The sentence travelled inside hours. Yahoo News / Business Insider ran the line as its headline — verbatim, in quote marks — and Business Insider's social post reproduced the formulation as its lede. The headline placement converted what would otherwise have been a piece of protest rhetoric into a recurring frame for the wider AI-regulation news cycle.

Propagation

The framing has propagated along three legible tracks.

Inside the AI-safety / pause movement, it became the campaign's working short-form for the regulatory-gap argument. PauseAI's own post-action Substack report on the DeepMind protest treats the line as one of the day's load-bearing artefacts alongside the gavel-and-jury mock trial, the Joep Meindertsma podium remarks, and the "Test, don't guess" chant. Third-party documentation in the AI-safety community has logged it the same way: the EA Forum's July 2025 overview of recent international demonstrations against AI lists "AI companies are less regulated than sandwich shops" as the working framing for the PauseAI protest line. The cognate formulations across the same protest cycle — including a Congressional AI lobbyist at the 4 July 2025 follow-on action using "they should be able to have the same regulation a meat packer or a hairdresser has" — illustrate the broader pattern: a single low-stakes industry held up against frontier-AI labs, with the asymmetry doing the rhetorical work.

In philanthropic and AI-safety communications, the framing was taken up almost immediately by the Future of Life Institute. By the 3 December 2025 launch of FLI's Winter 2025 AI Safety Index, Euronews's coverage was running with "AI 'less regulated than sandwiches' and no tech firm has AI superintelligence safety plan, study says" as its lede — and FLI president Max Tegmark was on record with the sandwich-comparator variant as part of the report's launch framing. By November 2025 the line had settled into Tegmark's stump rhetoric in long-form debate, where his debate-opening with Dean Ball on the proposed superintelligence ban took "right now there are more regulations on sandwiches than superintelligence in the US" as its opening sentence.

In adjacent movements, the framing has begun crossing into labour-side AI advocacy. The AFL-CIO's March 2026 use of the line — "AI is less regulated than sandwich shops. That's a problem." — applies the framing to the US labour federation's worker-side AI-regulation messaging, an early indicator of the kind of cross-movement uptake (from AI-safety / pause organising into the labour-side AI advocacy space) that the corpus's own creative-professions framing has separately demonstrated in the other direction.

Why it has carried

Three features have made the framing durable. First, it argues by analogy from a familiar regulatory regime — local food-safety inspection — that virtually every reader has direct experience of, and against an unfamiliar one (frontier-AI deployment) that most readers do not. The asymmetric familiarity is the point: the framing recruits the reader's existing trust in food-safety regimes as the evidentiary baseline, and lets the gap to AI do the rhetorical work. Second, it is morally weight-neutral on the sandwich industry — the framing carries no claim that sandwich shops are over-regulated — which keeps the argument focused on the AI side rather than collapsing into a deregulatory debate the AI-safety field has no interest in opening. Third, the framing is portable across registers. PauseAI uses it at the protest line; FLI uses it in a launch press release for a peer-reviewed-style safety index; Tegmark uses it in long-form policy debate; the AFL-CIO uses it in labour-federation social communications. Each register keeps the analogy intact while attaching it to different evidentiary bases — Frontier AI Safety Commitments and Seoul Summit follow-through for PauseAI's DeepMind protest, corporate AI-safety planning gaps for FLI, regulatory-capture and worker-protection arguments for the AFL-CIO.

The framing's limit, visible in the year since it entered the press cycle, is that the food-safety regulatory regime it invokes is itself contested in US deregulatory politics; subsequent uses of the framing (and of its cognates — meat packers, hairdressers) have had to navigate that adjacent debate without letting the AI argument get pulled into it. That navigation has been read across the movement as further evidence of the framing's durability: it has carried through a year of US regulatory churn without losing the structural argument it was built around.