Day: May 4, 2026

South Korea’s AI Basic Act, effective Jan 22, 2026, establishes a multi-layered regulatory framework that combines a binding horizontal AI law, cross-cutting data-protection rules, and sector-specific regulations. It imposes differentiated obligations—such as transparency for generative AI, risk management for high-impact AI, and enhanced safety for high-performance AI—while providing a one-year grace period for organizations to build compliant AI governance.

Agentic AI can autonomously take actions on behalf of users, raising legal, governance, and security risks that require clear authority definitions, human oversight, and robust monitoring. Organizations should implement pre‑deployment risk assessments, assign accountable owners, and establish controls such as audit logs and the ability to pause or disable agents.

Merz’s push to loosen EU AI rules for industrial sectors faced opposition from his coalition partner, Germany’s Social Democrats, who warned that the cuts would weaken consumer protections and favor foreign companies. The Social Democrats sent an urgent letter urging EU lawmakers to resist any changes that would undermine the AI Act’s horizontal approach.

The University of Johannesburg launched the “Beyond the Code: AI and Law” podcast, positioning the institution at the forefront of global AI governance debates. Hosted by Vice‑Chancellor Prof Letlhokwa George Mpedi, the series calls for enforceable regulations to ensure AI serves humanity responsibly.

The EU’s AI Act Omnibus negotiations have stalled, leaving the high‑risk AI system enforcement deadline of 2 August 2026 unchanged while debates continue over carve‑outs for sector‑specific regulations. A new trilogue is expected in the coming weeks, with the Irish EU presidency taking over on 30 June, which could shape the final path forward.

SAS unveiled a suite of AI governance tools, including the SAS AI Navigator and new Viya assistants, to make agentic AI trustworthy and compliant for enterprises. The company also announced upcoming initiatives like the SAS Quantum Lab, aimed at democratizing quantum AI while emphasizing human oversight.

EU AI Act reform talks have stalled, delaying the agreed postponement of high‑risk AI compliance dates and risking a return to the original 2 Aug 2026 deadline. Negotiators plan to resume discussions in two weeks, while critics warn the deadlock could increase compliance costs and fragment the regulatory framework.

The article argues that current knowledge about AI is insufficient for effective regulation, emphasizing the need for an industry consortium to develop flexible standards before formal government rules can be enacted. It highlights recent concerns, such as Anthropic’s Mythos model exposing zero‑day bugs, as examples of why cautious, collaborative oversight is essential.

AI safety faces a collective-action problem where competitive pressure drives labs to cut safety measures, and existing regulatory approaches struggle with information asymmetry, rapid pacing, and irreversible harms. Adopting a supervised self-regulatory organization, modeled on finance’s SROs, could coordinate standards, enforce rules, and provide real-time oversight while balancing industry expertise with public accountability.

The 2026 China Internet Civilization Conference in Nanning will focus on AI, releasing new Artificial Intelligence Application Ethics and Safety Guidelines. Over two days, the event features a main forum and 14 sub-forums addressing internet development, digital society, and responsible technology use.