top of page
Search

AI Regulation Retreats (June 2026)

  • Gov+AI
  • Jun 3
  • 5 min read

The dominant story in AI governance this week is not a new law. It's a delay. Across multiple jurisdictions, governments are pulling back timelines, softening obligations, and reframing regulation as a barrier to innovation rather than a safeguard against harm. For Australia, where the National AI Plan relies on voluntary frameworks and sector regulators to carry most of the governance load, what happens when those global frameworks weaken matters directly — because Australian businesses, consumers, and regulators all take cues from international standards, and because companies operating here also operate under these offshore regimes.


The Headline: Europe Rewrites Its Own Rulebook

The biggest development this week is the EU AI Act Digital Omnibus agreement, reached on 7 May 2026 between the European Parliament and Council. This is the first set of amendments to the EU AI Act since it was adopted in 2024. The amendments were driven by sustained industry lobbying — particularly from smaller technology companies and EU member states with large tech sectors — arguing that the original timelines were unworkable and that compliance costs would fall disproportionately on SMEs. The result is a package that extends deadlines, creates new carve-outs, and reduces some obligations, while adding two new prohibitions in high-harm areas.


The headline changes: compliance deadlines for high-risk AI systems (stand-alone systems in areas like employment, education, and law enforcement) have been pushed from August 2026 to December 2027 — a 16-month delay. High-risk AI embedded in regulated products like medical devices gets an additional year, now due August 2028. With the rate of tech advances, its hard to imagine what these regulations will actually be oversighting in over two years.  SMEs also gain new concessions: simplified documentation requirements, reduced fine caps, and priority sandbox access. The European Commission also published draft guidelines on classifying high-risk AI systems on 19 May, with a public consultation open until 23 June.


Some increased oversight has been achieved. Two new prohibited AI practices have been added to Article 5: the use of AI to generate or manipulate non-consensual intimate images, and child sexual abuse material. These prohibitions take effect December 2026 — moving faster than the high-risk compliance regime.


The amendments, as analysed by White & Case and Inside Privacy, represent a deliberate trade-off: extended timelines and reduced obligations for industry, while protecting the framework's overall architecture. These timeframe adjustments are not uncommon in new and emerging legislative frameworks, but should accompany careful scrutiny and monitoring to ensure the delay does not normalise non-compliance or reduce industry preparedness when deadlines do arrive.


Key Developments

United States — Federal Preemption, No Federal Law

The US continues its strange governance paradox: the Trump administration is aggressively pushing to preempt state AI laws through a National Policy Framework released in March, while simultaneously refusing to enact a standalone federal AI statute. The framework's six priorities — child safety, energy, intellectual property, anti-censorship, workforce, and lighter-touch regulation — signal an innovation-first posture that regards state-level consumer protection rules as obstacles.


This week the Colorado story sharpens that dynamic. Governor Polis signed SB 26-189, which substantially rewrites Colorado's landmark AI Act, delays its effective date to January 2027, and strips out its most robust provisions — including the duty of care to prevent algorithmic discrimination, impact assessment requirements, and proactive reporting to the attorney general. What remains is a narrower disclosure-and-transparency framework, enforced only by the attorney general. A federal court also stayed enforcement of the original law in late April. Colorado was meant to be a model for other states. It's been defanged.


United Kingdom — Sandboxes, Not Statutes

The UK confirmed this week there will be no standalone AI bill in the 2026 session. Instead, the government's "Regulating for Growth Bill" introduces sandbox powers to test AI deployment in healthcare, transport, and professional services. The UK also remains stuck on copyright and AI, with the March 2026 report confirming there is no preferred option on how to handle AI training on copyrighted material — a question with direct implications for Australian content creators and publishers.


Australia — Committed to Voluntary, For Now

Australia's National AI Plan, released in late 2025, remains the operative framework: voluntary guidance, sector-specific updates to existing laws, and an AI Safety Institute rather than a standalone Act. In practice this means Australia is relying on the Privacy Act, consumer protection laws, and instruments like the NSW Work Health and Safety Amendment (Digital Work Systems) Act 2026 — which imposes specific duties on employers using AI to allocate work — to carry the regulatory load. Australia also rejected a text-and-data-mining copyright exemption for AI training in April and is now exploring a licensing model — but there's no timeline.


China and the Indo-Pacific

China's amended Cybersecurity Law, effective January 2026, formally incorporates AI ethics and risk governance into its legal framework. Requirements for labelling AI-generated content in e-commerce came into force in February. South Korea and Vietnam also have new AI laws effective in 2026. The Indo-Pacific regulatory picture is fragmenting along national lines, which creates real compliance complexity for Australian businesses operating regionally.


What Does This Mean?

The pattern across tthese developments is consistent: compliance deadlines are being extended and substantive obligations are being reduced, driven primarily by industry pressure rather than evidence that original requirements were technically unachievable. The EU's 16-month delay is the most consequential example, but Colorado removing its algorithmic discrimination duty of care and the UK deferring legislation in favour of sandboxes reflect the same dynamic.


For Australia, this context matters for two reasons. First, many multinational companies operating in Australia are also subject to the EU AI Act — and delays in that framework reduce the pressure on those companies to embed AI governance practices globally, including in their Australian operations. Second, Australia's own National AI Plan relies heavily on voluntary compliance and the effectiveness of sector regulators. If international regulatory floors are lowering, the case for Australia moving toward more binding domestic obligations becomes stronger, not weaker — particularly for high-risk applications in healthcare, financial services, and employment.


The EU's new prohibitions on AI-generated non-consensual intimate images demonstrate that targeted, harm-specific regulation remains a priority and can move quickly when political will exists. That model — narrow, clearly scoped, enforceable rules in areas of demonstrated harm - appears to have a higher political apetite than broad outcomes or impact based regulation.


Summary

These developments share a common thread: compliance obligations are being reduced or deferred in response to industry pressure, while enforcement remains limited to the narrowest categories of harm. This reflects a broader political environment in which AI governance is being subordinated to economic competitiveness arguments. Australia's voluntary-led approach was designed for a world in which international frameworks provided a meaningful floor. As those frameworks soften, that assumption deserves reexamination.


Sources

[2] EU agrees Digital Omnibus deal to simplify AI rules – White & Case https://www.whitecase.com/insight-alert/eu-agrees-digital-omnibus-deal-simplify-ai-rules

[3] Artificial Intelligence: Council and Parliament agree to simplify and streamline rules – EU Council https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/

[4] European Commission Releases Draft Guidelines on High-Risk AI Under the EU AI Act – Hunton Andrews Kurth https://www.hunton.com/privacy-and-cybersecurity-law-blog/european-commission-releases-draft-guidelines-on-high-risk-ai-under-the-eu-ai-act

[5] President Donald J. Trump Unveils National AI Legislative Framework – The White House https://www.whitehouse.gov/releases/2026/03/president-donald-j-trump-unveils-national-ai-legislative-framework/

[7] AI in the King's Speech 2026: Regulating for Growth Bill announced – Bird & Bird https://www.twobirds.com/en/insights/2026/ai-in-the-kings-speech-2026-regulating-for-growth-bill-announced

[8] Australia's National AI Plan: big ambitions, but light on details – White & Case https://www.whitecase.com/insight-alert/australias-national-ai-plan-big-ambitions-light-details

[9] Australia launches National AI Plan – Department of Industry, Science and Resources https://www.industry.gov.au/publications/national-ai-plan

[10] Notes from Asia-Pacific: China's Cybersecurity Law amendments introduce AI provisions – IAPP https://iapp.org/news/a/notes-from-the-asia-pacific-region-china-s-cybersecurity-law-amendments-introduce-ai-provisions/

[12] EU AI Act omnibus: what changed on 7 May 2026 – VerifyWise https://verifywise.ai/blog/eu-ai-act-omnibus-what-changed

This blog was produced with assistance from AI. All sources have been verified.

 
 
bottom of page