Momentum builds for hard AI rules
Policymakers and experts say voluntary AI guardrails are running out of credibility and momentum is growing for enforceable, government-led regulation — especially for high‑stakes sectors like finance, health and security argued. Researchers warn development cycles are outpacing scientific oversight reported, and Indigenous groups in Canada are flagging harms from unchecked AI-generated cultural and language content reported.
The EU moved from promises to penalties when it began applying the AI Act’s first restrictions on Feb. 2, 2025 ([dlapiper.com)], with the law’s full “high‑risk” compliance framework scheduled to apply on Aug. 2, 2026 ([europarl.europa.eu)] and administrative fines reaching as high as €35 million or 7% of global turnover for the most serious breaches ([ai-act-service-desk.ec.europa.eu)]. The U.S. federal approach has shifted toward a single national standard after President Trump signed an Executive Order titled “Ensuring a National Policy Framework for Artificial Intelligence” on Dec. 11, 2025 directing federal action to preempt state laws ([federalregister.gov)]; at the same time the SEC held an industry roundtable on March 27, 2025 and flagged AI as an exam priority for FY2026, signaling tougher enforcement on disclosures and recordkeeping in finance ([natlawreview.com)], while the New York Department of Financial Services issued AI‑cyber guidance for banks and insurers on Oct. 16, 2024 ([dfs.ny.gov)]. High‑profile industry pledges have scrambled: major firms agreed to White House voluntary commitments on July 21, 2023 but independent analysis found limited transparency and accountability since then ([technologyreview.com)], and Anthropic’s publication of Responsible Scaling Policy v3 on Feb. 24, 2026 rewrote its earlier “hard pause” pledge—prompting coverage that the company dropped a flagship safety limit ([anthropic.com)]. Researchers and methodologists warn oversight is chronically slow: a Science/Cambridge‑linked analysis argues empirical work on digital harms “moves too slowly” to hold tech to account ([cam.ac.uk)], and commentary tracking AI research timelines finds peer‑review and policy cycles lagging behind deployed systems, with monitors like chain‑of‑thought transparency increasingly fragile ([linkdood.com)]. Indigenous communities in Canada have publicly flagged concrete harms from AI‑generated cultural and language content, with CBC reporting AI‑generated dictionaries and elders’ teachings circulating in ways that can undermine revitalization efforts ([thenote.app)]; Canada’s federal AIDA legislative bid stalled when Bill C‑27 died on Jan. 6, 2025, and Ottawa published a national AI strategy engagement summary on Feb. 2, 2026 as it seeks alternative policy routes ([osler.com)]. Markets and compliance functions are already reacting: Gartner projects AI governance spending at about $492 million in 2026 and expects it to exceed $1 billion by 2030 as fragmented rules drive demand for audit, monitoring and risk tools ([gartner.com)], while industry surveys show compliance officers and finance leaders ranking AI oversight as a top priority heading into 2026. ([acaglobal.com)]
