Show the Working

Drive Innovation, Regulate AI, and Support Diverse Business Models

Labour · what the evidence says

An independent, source-checked look at Labour’s policy “Drive Innovation, Regulate AI, and Support Diverse Business Models” — what it would actually do across the things that affect your life. Every claim below quotes the source behind it. How this works.

Personal liberty & free speech — Mixed picture

minor · moderate confidence

The policy creates new criminal offences around deepfakes and binding AI regulation that restrict certain expression and activity, while the Regulatory Innovation Office trims some state-imposed barriers for businesses. The net liberty effect is small and pulls in opposite directions.

The evidence

Biggest unknown: How broadly the 'binding regulation on powerful AI models' is ultimately drafted — a narrow, targeted regime has modest liberty costs; a sweeping one could chill research and expression significantly.

Our reading: This policy has two countervailing O10 effects. On the worsening side: the deepfake ban creates a new criminal offence covering a category of content production — this is an unambiguous new state restriction on expression, even if the targeted content is harmful. The binding AI regulation imposes legal obligations on developers of powerful models; while narrowly scoped per the evidence, it constitutes new state coercion over a class of actors. The NDL consolidates government access to public-sector data including potentially health data; the evidence flags genuine public concern about privacy, though 'robust safeguards' are promised. On the improving side: the RIO explicitly aims to reduce regulatory red tape for startups and high-growth firms, removing some existing state-imposed barriers. This is a real, if limited, reduction in regulatory coercion for businesses. On balance, the liberty costs are more concrete (criminalization of deepfakes is a definitive new restriction; NDL raises genuine privacy risk) than the liberty gains (RIO deregulation is real but narrow and implementation-dependent). Neither side dominates overwhelmingly — the deepfake ban is targeted and addresses genuine harms, the AI regulation is narrow, and the RIO deregulation is genuine — so 'mixed/minor' best captures this. The biggest swing factor is the final AI regulation's breadth, which remains unspecified.

Prosperity & living standards — Helps

moderate · low confidence

This package of AI, R&D, and data policies could meaningfully boost UK productivity and economic opportunity over the long run, but most benefits depend on implementation details that are not yet settled, and the co-operative doubling target faces serious structural barriers. Near-term effects on living standards are likely modest.

The evidence

Biggest unknown: Whether the Regulatory Innovation Office, National Data Library, and ten-year R&D budgets will be implemented with sufficient scale and coherence to fire the productivity mechanism at population level, or remain well-intentioned but fragmented initiatives.

Our reading: This policy bundle targets three channels that could lift O13: (1) AI and data infrastructure investment, (2) long-term R&D stability, and (3) expansion of more productive business models via co-operatives. On the AI/data side, removing planning barriers for datacentres accelerates infrastructure that industry bodies identify as demand-driven by AI growth, and the NDL carries projections of very large long-run societal gains — though those projections come from advocacy-adjacent sources and implementation is incomplete. Ten-year R&D budgets address a real structural weakness: instability in research funding undermines private sector confidence and talent retention. This is positively received by independent voices (CaSE, techUK), though coverage is targeted rather than universal. The Regulatory Innovation Office is welcomed in principle but analysts raise legitimate concerns that it may layer rather than simplify regulation. The co-operative doubling target has some genuine productivity evidence behind it — higher labour productivity, better survival rates — and plausible mechanisms. But achieving 7.2% annual growth against structural finance and legal barriers, with only a call for evidence launched as of late 2025, lacks a committed delivery instrument sufficient to earn 'improves' on that strand alone. Taking the bundle together: the AI infrastructure, data, and R&D stability measures have plausible, moderately evidence-supported mechanisms for long-term productivity gains; the co-operative target is ambitious but under-mechanised. Absent this policy, UK AI infrastructure would face continued planning friction, R&D institutions would face short funding horizons, and co-operative sector growth would remain constrained. The marginal gain is real but back-loaded and conditional on delivery. Direction: improves, long-term, moderate — but confidence is low given implementation gaps across all three strands.

Inequality & fair shares — Little effect

minor · low confidence

The policy's biggest potential lever on inequality — doubling the co-operative and mutuals sector — could broaden ownership and narrow the wealth gap, but the concrete mechanism is limited to a call for evidence. The AI and R&D measures lack any cited distributional analysis.

The evidence

Biggest unknown: Whether the government will introduce the specific finance, legal, and regulatory reforms needed to actually grow the co-op sector at the required rate — without those levers the commitment is aspirational.

Our reading: O14 asks whether the gap between richest and rest narrows or widens. The most relevant plank of this policy for inequality is the commitment to double the co-operative and mutuals sector: broader employee and community ownership is a recognised mechanism for narrowing wealth inequality, since co-ops distribute ownership and surpluses more widely than conventional shareholder firms. If the sector doubled, the distributional effect could be meaningful. However, applying the soft-verb and mechanism-plausibility rules, the commitment stalls at aspiration. The only concrete government action cited is a call for evidence (E37), while analysts identify substantial structural barriers — finance access, legal frameworks, investor awareness — that remain unaddressed (E36). Achieving the required 7.2% annual growth rate, double the wider economic forecast, without those instruments is not evidenced as plausible. On the AI, datacentre, and R&D side, none of the provided evidence units contain any distributional or inequality analysis: the benefits cited are aggregate (GVA, innovation, cost savings) and the beneficiaries appear to be large tech firms and investors, but no evidence directly addresses the gap between top and bottom. Without a cited distributional study, asserting that AI infrastructure spending widens inequality would be importing ungrounded reasoning. The verdict is therefore negligible: the co-op element points in an inequality-narrowing direction but lacks a committed instrument to fire at scale, and the remaining measures have no cited distributional signal at all. Confidence is low because the co-op mechanism is theoretically sound but evidentially undelivered, and the AI/R&D elements are simply unscored for distribution in the provided evidence.

Good work & fair pay — Mixed picture

minor · low confidence

This policy bundle could improve job quality and pay over time through AI sector growth, co-operative expansion, and more stable R&D funding — but the biggest gains depend on delivery of ambitious targets like doubling the co-op sector, which faces serious structural barriers. The near-term effect on ordinary workers' pay and security is likely small.

The evidence

Biggest unknown: Whether the co-operative doubling target and AI sector growth actually translate into more secure, better-paid jobs at scale, or whether barriers to co-op finance and regulatory implementation stall delivery.

Our reading: This policy bundle touches O4 through three channels: AI sector investment, co-operative sector growth, and R&D stability. On the co-operative channel, the evidence shows the sector already supports ~4 million jobs and that worker co-ops outperform traditional firms on productivity and survival. If the doubling target were achieved, it would plausibly improve job security and pay quality for a material share of the workforce. However, achieving 7.2% annual growth — twice the projected economy-wide rate — in the face of documented barriers around finance access, outdated legal frameworks, and limited institutional support makes this a highly ambitious projection. The call for evidence only began in late 2025, meaning concrete instruments are not yet in place. This limits the verdict to 'projected' gains rather than a confident 'improves'. On the AI/R&D channel, ten-year budgets could improve talent retention and create more stable, skilled employment in priority sectors. The RIO might reduce friction for high-growth tech businesses, supporting job creation — but credible analysts note it may add regulatory complexity rather than reduce it, and the delivery risk is real. Binding AI regulation is narrowly scoped and primarily a safety/rights measure; its direct effect on wages or employment security for ordinary workers is minimal. The deepfake ban addresses workplace harassment risk (relevant to job quality), but its effect on O4 at population scale is marginal. Overall, the policy points in the right direction for work quality — co-op expansion and R&D stability both correlate with better jobs — but the mechanisms are either aspirational (doubling target without funded instruments) or long-horizon (R&D culture change). The counterfactual absent this policy is modest organic sector growth, so the marginal gain is real but uncertain and distant. 'Mixed/minor' reflects genuine upside potential tempered by serious delivery uncertainty and the absence of near-term funded mechanisms.

Crime, justice & national security — Helps

minor · low confidence

Banning sexually explicit deepfakes and introducing binding safety rules for powerful AI models should reduce a specific and fast-growing category of harm. The overall effect on crime and security is real but narrow, and delivery depends on how quickly and robustly the legislation is enforced.

The evidence

Biggest unknown: Whether the Crime and Policing Bill's deepfake criminalisation will be enforced effectively enough to deter a harm that is growing rapidly and often crosses international jurisdictions.

Our reading: The policy's most direct contribution to O5 is the criminalisation of sexually explicit deepfakes. This is a concrete legislative commitment — not an aspiration — with a named delivery vehicle (Crime and Policing Bill). The baseline evidence shows this is a rapidly growing harm (500k to 8 million files in two years) used for harassment, and existing law is acknowledged as inadequate. A specific criminal offence creates a new deterrent and enforcement tool, which is a genuine improvement to the justice system's coverage of this harm, even if modest in the wider crime picture. The binding AI safety regulation for frontier models also has a plausible O5 pathway: powerful AI systems pose national security and public-safety risks, and targeted binding rules go beyond the current soft-law landscape. However, the specific regulatory content remains undefined, so its safety impact cannot yet be quantified. The absence of detail, combined with the narrow scope (a handful of companies), means this element earns only candidacy for an improvement rather than a confident verdict. Absent this policy, deepfake harassment would remain covered only by general harassment law, which experts flag as insufficient. The deepfake ban is genuinely additional. The overall effect is real but bounded — it addresses one specific and growing harm rather than crime or security broadly. Confidence is low because the deepfake ban has not yet passed into law and enforcement capacity is unproven, while the AI safety regulation lacks published substance.

Equal treatment & democratic rights — Helps

minor · moderate confidence

Banning sexually explicit deepfakes gives real legal protection to people — mostly women — who face this form of AI-driven harassment, which is a genuine equal-treatment gain. The rest of the policy bundle touches O9 only marginally, if at all.

The evidence

Biggest unknown: Whether the Crime and Policing Bill is enacted as described and whether enforcement proves effective against anonymous or overseas creators of deepfakes.

Our reading: Of the five policy strands, only the deepfake ban makes a clear and direct contact with O9. Sexually explicit deepfakes are a documented, fast-growing form of targeted harassment; criminalising them is a concrete legal mechanism — a named Bill — rather than an aspiration, satisfying the threshold for 'improves'. The harm is real at scale (8 million files estimated by 2025) and disproportionately affects women and other targeted groups, making this an equal-treatment protection. Some gap exists because existing harassment law already partially covers such conduct, so the marginal gain is real but not transformative; hence minor rather than moderate magnitude. The binding AI regulation is stated but underspecified: without knowing what obligations it imposes, it cannot be scored as an equal-treatment improvement. The National Data Library raises questions of data governance and consent (E10), but these land more naturally on O10 (privacy/liberty) than O9. The Regulatory Innovation Office, R&D budgets, and co-operative sector doubling have no plausible direct pathway to equal treatment or democratic rights under the strict O9 scope. Overall the verdict is a modest but genuine improvement, conditional on the Crime and Policing Bill passing and on enforcement being practically feasible — especially against offshore or anonymous creators.

Compare all parties on personal liberty & free speech →Compare all parties on prosperity & living standards →Compare all parties on inequality & fair shares →Compare all parties on good work & fair pay →Compare all parties on crime, justice & national security →Compare all parties on equal treatment & democratic rights →Open this in the interactive comparison →← Show the Working home