Do EU delays of AI products line up with the AI Act's dates or with other EU rules?

Research question

For AI products and features released from 2023 onwards by a broad set of providers (not only frontier labs), how often is availability in the EU delayed or withheld relative to the US, the UK and Switzerland, and do delays cluster around the dates when the AI Act's general-purpose model obligations applied (August 2025) and became enforceable (August 2026), separately from data-protection and competition-law events?

Why it matters

Claims that EU law keeps AI products out of Europe are central to policy fights over simplification and delay, but they rarely separate one law from another or a law's announcement from its application. A design that ties delays to legal stage would show whether delays line up with the AI Act's own dates rather than with data-protection enforcement or competition law. It shows association with dated events, not proof that a law caused a delay.

The debate lists this among what we still do not know: causal product-withholding effects attributable to a specific EU AI law

Evidence that motivates it

1. The EU AI Act is in force, but Annex III high-risk obligations apply from 2 December 2027 and Annex I product-embedded high-risk obligations from 2 August 2028. (LD12-HQ-C01 v1.0)
   The Act is in force but its high-risk obligations apply only from December 2027 (Annex III) and August 2028 (Annex I), which gives the dated legal events this design exploits; the claim does not license treating the Act as inactive.
   Cites: Regulation (EU) 2026/1744; European Commission AI Omnibus…
2. Italy's €15 million GDPR fine on OpenAI was annulled by the Court of Rome in March 2026 on competence grounds under the GDPR one-stop-shop; the court did not resolve the substantive allegations. (LD12-HQ-C09 v1.0)
   Data-protection enforcement against generative AI is active but contested (the Italian fine on OpenAI was annulled on competence grounds), so it is a competing explanation for withholding that the design must separate from the AI Act.
   Cites: Garante Decision No. 755 of 2 Nov 2024; Court of Rome R.G. 4785/2025 judgment

Claims are shown as they read on the day this pack was built; claim pages keep every earlier version.

A possible design

Extend an existing release-delay dataset in two directions. (1) Coverage: add product and feature releases from 15–25 providers including European, Chinese and mid-sized firms and AI features inside consumer products (assistants, image and voice tools), recording first availability by jurisdiction from release notes, help-centre country lists and archived pages. (2) Attribution: code each delay's stated reason from provider statements and press, and separately estimate an event-study of the share of releases delayed in the EU relative to the UK and Switzerland around dated events: general-purpose model obligations applying (August 2025) and becoming enforceable (August 2026), high-risk dates, major data-protection decisions, and Digital Markets Act decisions. Treat provider statements as claims to be checked, not as causes. Pre-register the event list and the comparison countries.

Data needed

Release notes, country-availability pages and Internet Archive snapshots; an event calendar of EU legal stages and enforcement decisions; the existing GovAI dataset as a base if its authors share it.

Feasibility

One PhD chapter or paper. Low: researcher time; possibly a research assistant for data collection. Feature-level availability is messy and many delays have several causes, so results should be reported as associations with dated events plus coded stated reasons, not as clean causal estimates. Switzerland and the UK have their own rules, which must be logged as events too. A study run now can cover releases to late 2026 and the two general-purpose model dates; the high-risk obligations apply from December 2027 (Annex III) and August 2028 (Annex I), so testing those dates needs a later extension of the same dataset.

Has this been done?

Partly done already: related work covers some of this question (first-pass screen of 12 October 2026, checked again on 13 October 2026).

GovAI's 2026 report covers 375 model releases (text models only) from four US frontier labs to May 2026 and tentatively attributes most delays to data-protection regulation (only five attributions are high-confidence), finding no strong evidence that the AI Act caused delays but noting its data end before the general-purpose provisions became enforceable; it uses no event-study design and no Switzerland comparison. An industry-funded survey reports firm-stated delays, and policy commentary lists examples of features withheld from the EU (including 2026 Apple and Google features linked by the firms or commentators to the DMA) without a dataset. What remains open is a broader set of providers, modalities and product features, coverage past August 2026, and a design that separates the AI Act's legal stages from data-protection and competition-law events.

Literature check. Reading the full GovAI report (Lidiard et al. 2026) confirms it covers model releases from four US labs to May 2026, excludes non-text models, uses no event study, and does not use Switzerland as a comparison; its authors list startups, non-US firms, app builders and other modalities as future work. Seven new queries (preprint, economics-journal, forward, feature-tracker and German-language searches) found no other study that ties EU availability delays to dated AI Act, data-protection and DMA events: what exists is commentary listing examples (CEPA, Nov 2024; Chamber of Progress, Jun 2026; a VoxEU column, Aug 2026) and the GDPR app-exit literature as a methodological model. The opportunity stands as written. Checked by Claude (SocioLens Project Lead) with AI research agents, by web search; not a person and not a domain expert.

Where the search could not reach. Screened by web search on the date shown; SSRN, Google Scholar and paywalled economics journals were only partly reachable, and recent working papers in economics may be missed. Absence of a hit is not proof that the work does not exist. Literature check of 2026-10-13: Economics working papers on SSRN and CEPR discussion papers could not be searched through their own interfaces, and Google Scholar was not searched, so a 2026 working paper could be missed. Commentary pieces were read in full from their pages; the GovAI report was read in full in an earlier pass this round. Absence of a hit is not proof that the work does not exist.

Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.

Nearest work found

• Lidiard, Vereschak, Gibbs & Anderljung, "Delays to Frontier AI in the EU and UK," GovAI Technical Report (28 Jun 2026)
  375 releases from Meta, Google, OpenAI and Anthropic (2018–May 2026): 11% delayed or not released in the EU and 7% in the UK; 56 of 68 delays tentatively attributed mainly to regulation, chiefly data protection; no strong evidence of AI Act effects. Attribution is by judgment from public sources; four US labs only; not peer reviewed. The first screen used only the report's landing page; the full report was read in the literature check of 13 October 2026.
• Janßen, Kesler, Kummer & Waldfogel, "GDPR and the Lost Generation of Innovative Apps," NBER Working Paper 30028 (May 2022) Judged from an abstract, summary or search listing only
  Shows how a single EU law's start date can be linked to app exit and entry on Google Play (about a third of apps exited; entry fell sharply). A methodological model for dating regulatory effects; concerns the GDPR and apps generally, not AI or the AI Act. Read from the NBER abstract page.
• Association for Competitive Technology, "The Hidden Cost of AI Regulations for EU and UK Startups and SMEs" (9 Oct 2025)
  Commissioned surveys of more than 1,000 small technology firms in the EU, UK and US; six in ten EU/UK firms reported delayed access to frontier models and 58% of EU/UK developers reported regulation-driven delays. Published by an industry advocacy group; sampling method not described; firm-stated causes, no attribution to a specific law.
• Perugini, "Europe's regulatory double bind," VoxEU/CEPR column (14 Aug 2026)
  Interpretive column on the Digital Omnibus delay of high-risk obligations, missing harmonised standards and certification timelines. Cites no dataset on AI launch timing and attributes no delays to specific laws quantitatively; states that the causal weight of external pressure versus implementation constraints cannot be determined from public sources. Context only.
• Lagercrantz, "Europe's AI Blues: US Companies Slow Deployment," CEPA (1 Nov 2024)
  Commentary listing named cases (Google AI Overviews, Bard, Meta multimodal Llama and AI training, Apple Intelligence; OpenAI Advanced Voice Mode as a counterexample) and attributing them to GDPR, the DMA or general uncertainty from company statements and press. No count, dataset or causal design; useful as a source of candidate events and stated reasons.
• Jebelli, "Digital Markets Act Delays Apple & Google AI in Europe," Chamber of Progress (8 Jun 2026)
  Industry-advocacy opinion piece listing 2026 features withheld or US-first (Apple Siri AI in iOS 27; Google Gemini Spark, Daily Brief, AI Inbox, Ask YouTube). No dataset or method; ties Apple's delay to the DMA via Apple's own statement and Google's only by argument. Shows the feature-level, post-August-2026 coverage the opportunity proposes is still unsystematised.

References

Sources behind the motivating claims:

• Regulation (EU) 2026/1744; European Commission AI Omnibus implementation material (SRC-LD12-EU-AI-ACT-2026-1744, one of the original documents; the reference cites several documents and this link opens only one)
• Garante Decision No. 755 of 2 Nov 2024 (SRC-LD12-GARANTE-DECISION-755-2024, original)
• Court of Rome R.G. 4785/2025 judgment (SRC-LD12-COURT-OF-ROME-RG-4785-2025, no checked link to the original yet)

BibTeX

The first entry cites this opportunity; then its nearest work; then the sources above. Nearest work and sources are recorded as reference strings, so each goes in whole as the title; add authors and year from the original before you cite it.

@misc{sociolens_ro_ld12_05_v3,
  author = {{SocioLens project}},
  title = {Do EU delays of AI products line up with the AI Act's dates or with other EU rules?},
  year = {2026},
  date = {2026-10-13},
  howpublished = {\url{https://sociolens-web.pages.dev/opportunities/ro-ld12-05/v3/}},
  note = {Research opportunity RO-LD12-05, version v3}
}

@misc{ro_ld12_05_nearest_1,
  title = {Lidiard, Vereschak, Gibbs \& Anderljung, "Delays to Frontier AI in the EU and UK," GovAI Technical Report (28 Jun 2026)},
  url = {https://www.governance.ai/research-paper/delays-to-frontier-ai-in-the-eu-and-uk},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld12_05_nearest_2,
  title = {Janßen, Kesler, Kummer \& Waldfogel, "GDPR and the Lost Generation of Innovative Apps," NBER Working Paper 30028 (May 2022)},
  url = {https://www.nber.org/system/files/working_papers/w30028/w30028.pdf},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only}
}

@misc{ro_ld12_05_nearest_3,
  title = {Association for Competitive Technology, "The Hidden Cost of AI Regulations for EU and UK Startups and SMEs" (9 Oct 2025)},
  url = {https://actonline.org/the-hidden-cost-of-ai-regulations-a-survey-of-eu-uk-and-u-s-companies/},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld12_05_nearest_4,
  title = {Perugini, "Europe's regulatory double bind," VoxEU/CEPR column (14 Aug 2026)},
  url = {https://cepr.org/voxeu/columns/europes-regulatory-double-bind},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld12_05_nearest_5,
  title = {Lagercrantz, "Europe's AI Blues: US Companies Slow Deployment," CEPA (1 Nov 2024)},
  url = {https://cepa.org/article/europes-ai-blues-us-companies-slow-deployment/},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld12_05_nearest_6,
  title = {Jebelli, "Digital Markets Act Delays Apple \& Google AI in Europe," Chamber of Progress (8 Jun 2026)},
  url = {https://progresschamber.org/insights/dma-delays-apple-google-ai-europe/},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{src_ld12_eu_ai_act_2026_1744,
  title = {Regulation (EU) 2026/1744; European Commission AI Omnibus implementation material},
  url = {https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; the link opens only one of the several documents this reference cites}
}

@misc{src_ld12_garante_decision_755_2024,
  title = {Garante Decision No. 755 of 2 Nov 2024},
  url = {https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/10085455},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{src_ld12_court_of_rome_rg_4785_2025,
  title = {Court of Rome R.G. 4785/2025 judgment},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}