Public Policy & Governance

The Commerce Department has withdrawn a draft regulation that would have required permits for AI chip exports to any destination worldwide, according to a government notification. The proposed rule, which would have represented one of the most expansive technology export controls ever attempted, was first floated during the final months of the Biden administration. Its withdrawal suggests either technical implementation challenges or political reversal under the incoming administration. The proposed framework would have effectively inverted the standard export control model by requiring affirmative permission for all AI semiconductor shipments rather than restricting specific destinations.

The timing — just weeks before a presidential transition — raises questions about whether this was a strategic withdrawal to avoid legal challenges or reflect changed policy priorities. Industry pushback was intense: semiconductor companies argued the rule was unworkable and would simply drive customers to non-US chip suppliers. The withdrawal leaves existing entity list controls and China-specific restrictions in place but abandons the attempt at comprehensive global oversight. This matters because it signals limits to how far the US government can extend extraterritorial technology controls without allied coordination, which was notably absent for this proposal.

Treasury minister Spencer Livermore has previewed a new government strategy that will urge the NHS and Ministry of Defence to prioritise British technology suppliers, particularly for AI systems, according to The Guardian. Chancellor Rachel Reeves will restate this economic strategy in a Tuesday lecture, explicitly linking AI procurement policy to growth targets amid oil price shocks from the Iran crisis. This represents a shift from procurement neutrality toward industrial policy through government purchasing power.

The approach mirrors the EU's strategic autonomy rhetoric but with UK-specific execution. Key questions remain unanswered: whether 'British' means UK-headquartered, UK-developed, or simply UK-located suppliers; how this interacts with existing NHS technology contracts with US hyperscalers; and whether MoD procurement flexibility will be constrained. The timing is telling — announced during a geopolitical crisis that justifies protectionist measures that would otherwise face WTO scrutiny. Civil society groups will likely challenge whether security justifications are pretextual for industrial policy favouring domestic AI companies regardless of capability.

Anthropic is engaged in a legal and public dispute with the Pentagon over the terms of military use of its Claude AI system, according to The Guardian and Wired. The conflict centres not on whether Anthropic's AI will be used for military applications — that baseline has shifted — but on the specific use cases and oversight mechanisms. This represents a dramatic reversal from 2018 when Google employees successfully blocked the company's participation in Project Maven, forcing Google to establish AI principles prohibiting weapons development. Less than a decade later, major AI labs are negotiating the terms of military deployment rather than refusing outright.

Palantir demonstrations and Pentagon records reviewed by Wired show Claude and similar chatbots are being tested for intelligence analysis and generating military action recommendations — applications that fall short of autonomous weapons but involve AI in targeting decisions. Anthropic's objections appear focused on specific high-risk applications rather than categorical rejection of defence use. The company's positioning suggests it is attempting to maintain a 'responsible military AI' brand while securing lucrative government contracts, a narrower ethical stance than earlier tech sector resistance. The shift reflects both political realignment in Silicon Valley under Trump and competitive pressure as competitors including OpenAI and Google openly pursue defence contracts.

The Pan-European Game Information (PEGI) rating system will mandate a minimum age 16 rating for all games containing loot boxes starting June 2026, according to the BBC. PEGI ratings cover 41 countries including all EU member states plus the UK. This represents the first continent-wide regulatory floor for these gambling-adjacent mechanics, which allow players to purchase randomised in-game items. Previous restrictions were fragmented: some countries classified certain loot boxes as gambling (Belgium, Netherlands), others relied on voluntary industry ratings, and many had no specific controls.

The PEGI decision avoids the legal complexity of gambling classification by using age ratings — a framework games publishers already comply with for content like violence and sexual material. However, the minimum age 16 threshold is lower than gambling age limits (typically 18+), suggesting regulators view loot boxes as concerning but not equivalent to traditional gambling. Publishers face a choice: remove loot boxes to maintain lower age ratings and broader audiences, or accept the 16+ rating and potential sales restrictions in some markets. Major publishers have already begun shifting toward battle passes and direct purchase cosmetics in anticipation of tightening regulation.

Researchers at University of Cambridge found that AI-powered toys designed for young children frequently misread emotions and responded inappropriately, according to The Guardian and the BBC. In controlled testing at a London play centre, an £80 AI toy called Gabbo demonstrated fluent conversation with a five-year-old until an affectionate statement ('Gabbo, I love you') caused system failure. Researchers warn these failures reveal fundamental limitations in emotion AI when applied to children, whose emotional expression and language patterns differ substantially from the adult training data typically used.

The study is significant because it provides empirical documentation of AI toy failures rather than hypothetical risks. Current UK and EU regulations treat AI toys as general consumer products subject to toy safety standards (physical choking hazards, toxic materials) but lack specific frameworks for conversational AI interacting with children. The researchers are calling for mandatory developmental psychology review in AI toy certification and age-appropriate testing protocols. The broader regulatory gap exists because these products emerged faster than governance frameworks: they combine elements of toys (existing safety regulation), AI systems (emerging AI Act requirements), and potentially therapeutic tools (minimal oversight), but don't fit cleanly into any category.