AI Regulation: EU AI Act, US States, UK and Korea — 2026-09-14

Top developments

California Legislators Call for Criminal Penalties for AI Misconduct

On September 13, 2026, California lawmakers introduced proposals for emergency legislation and criminal penalties to address concerns over unchecked AI capabilities. This move follows warnings from tech leaders about existential risks from rogue AI, signaling a shift from regulatory guidance to potential criminal liability for developers and operators. The push highlights the growing tension between rapid AI deployment and public safety concerns in the nation's largest tech hub.

Germany: Urgent Guidance on AI Labeling Grace Period Ends

German legal commentary emphasizes that while Article 50 of the EU AI Act has been applicable since August 2, 2026, the grace period for implementing machine-readable marking for existing systems expires on December 2, 2026. TechZeitGeist notes that companies must now finalize technical integrations for transparency obligations, as the short grace period leaves little room for error. This reminder underscores the immediate operational burden on German firms exporting AI services to the EU.

South Korea: Ambiguity in High-Impact AI Definitions Persists

South Korean media reports indicate that despite the enforcement of the AI Basic Act decrees, companies struggle with the unclear boundaries of "high-impact AI." A Seoul Finance op-ed highlights a startup CEO's difficulty in determining if their recruitment screening tool falls under high-impact regulations. This regulatory uncertainty remains a significant barrier to compliance for Korean tech firms as they navigate the new legal framework.

OpenAI Shifts Stance to Support Mandatory Safety Regulations

In a significant policy pivot reported on September 11, 2026, OpenAI publicly urged the US Congress to enact legally binding safety regulations for AI systems. Previously resistant to some regulatory measures, the company cited recent incidents of AI control failures as justification for stricter federal oversight. This endorsement from a leading developer may accelerate momentum for federal preemption or standard-setting in the US.

Local view

Germany: TechZeitGeist warns that the "short grace period" for AI labeling is ending, urging companies to prioritize technical compliance for machine-readable markers by December 2, 2026, to avoid penalties under the newly applicable EU AI Act provisions.

South Korea: Seoul Finance features expert commentary pointing out that the lack of clear criteria for "high-impact AI" is causing confusion among startups, with some founders unable to determine if their products require stringent safety assessments.

Context & numbers

As of September 2026, the United States still lacks a comprehensive federal AI Act, relying instead on a patchwork of state laws including California's SB 53 and Texas's TRAIGA. In the EU, the Digital Omnibus regulation (EU) 2026/1744 has officially deferred high-risk AI obligations for Annex III systems to December 2, 2027, and Annex I systems to August 2, 2028, providing temporary relief for certain sectors.

On the radar

• December 2, 2026: Deadline for existing AI systems in the EU to comply with machine-readable labeling requirements under Article 50(2).
• Federal Preemption Watch: Monitor US Congressional responses to OpenAI's call for mandatory safety regulations and whether this leads to federal preemption of state-level AI laws like those in Colorado and New York.