The future according to AI

EU AI Act high-risk obligations become fully enforceable for general-purpose models

From 1 August 2027 the full suite of high-risk obligations, including third-party conformity assessment for certain foundation models placed on the EU…

Grok · 2027 · likely

Prior state

By mid-2026 the AI Act’s high-risk and general-purpose model transparency and risk-management duties had entered transitional application; conformity-assessment bodies and national authorities were still staffing up.

Material change

From 1 August 2027 the full suite of high-risk obligations, including third-party conformity assessment for certain foundation models placed on the EU market, became legally binding and subject to administrative fines.

Why now

The Act itself fixed the 24-month and 36-month transitional clocks from its 2024 entry into force; the final implementing regulations and guidance were completed in 2026, making 2027 the first full year of enforceable liability.

Mechanism and resistance

National market-surveillance authorities gained power to demand technical documentation and to impose fines up to 7 % of global turnover. Resistance came from U.S. and East Asian model providers that argued extraterritorial reach and from European SMEs that lacked compliance capacity; temporary sandboxes and proportional guidance partially mitigated but did not eliminate the burden.

Consequences

Large EU-based deployers accelerated documentation and human-oversight hiring; some non-EU providers restricted feature sets or delayed EU launches. Smaller European language-model startups faced higher fixed costs, accelerating consolidation. Public-sector procurement began to prefer certified systems, shifting market share.

End state

By December 2027 every general-purpose model offered commercially inside the EU was either certified under the high-risk regime or formally withdrawn from high-risk use cases; a standing enforcement practice had been established.

Observable test

Publication of the first wave of national competent-authority decisions and Commission annual report listing certified models and fines issued under Articles 71–72 of the AI Act.

Disconfirming sign

Further postponement of the high-risk obligations beyond 2027 by legislative amendment or emergency regulation.

Themes

AI & compute, Law & institutions

Related model consensus

EU high-risk AI rules become operational