EU finalises AI Act; US gates GPT-5.6; emissions rise
The EU Council tightened the AI Act, the US coordinated GPT-5.6's rollout with government partners, and AI data-centre emissions climbed sharply.
By BINA Editorial
Today's stories share a common thread: governance frameworks are beginning to move at the same speed as AI capabilities.
EU Council Streamlines AI Act, Bans AI-Generated Intimate Imagery
The Council of the European Union gave its final approval this week to a package of amendments streamlining the AI Act's compliance requirements. The most significant change compresses the provider transparency deadline from six to three months, requiring affected companies to comply by 2 December 2026 — three months earlier than originally scheduled. The amendments also introduce an explicit ban on AI-generated non-consensual intimate imagery, also effective December 2026. Together, these changes sharpen Europe's most comprehensive AI governance framework ahead of full enforcement.
UN Scientific Panel Issues Its First Global AI Assessment
The UN Independent International Scientific Panel on AI released its inaugural assessment report, finding that global AI governance instruments remain fragmented and that access to AI's benefits is unevenly distributed — particularly in lower-income countries. The report is the first consolidated scientific review of AI risks and opportunities produced under a UN mandate at global scale. Its findings will be formally presented at the first UN Global Dialogue on AI Governance, which convenes in Geneva on 6–7 July.
US Government Coordinates GPT-5.6 Rollout — A First for Frontier AI
OpenAI unveiled the GPT-5.6 family — three models named Sol, Terra, and Luna — on 26 June, but made them available only to approximately 20 government-vetted partner organisations rather than the general public. Sol Ultra, the most capable variant, scored 91.9% on Terminal-Bench 2.1. The coordinated rollout is the first time a frontier model launch has been explicitly gated behind government review before any public release — a precedent that may define how the United States manages the deployment of the next generation of high-capability models.
Anthropic Makes Claude Sonnet 5 the Default; Fable 5 Returns After Export Ban Lifted
Anthropic has made Claude Sonnet 5 the default model for all users at standard Sonnet-tier pricing, delivering near-Opus performance at a lower cost tier. The company also restored global access to its Fable 5 model after the US Department of Commerce lifted the export control order that had blocked international availability since June. To accompany the broader rollout, Anthropic released a new safety classifier designed to block jailbreak attempts, which the company says is essential for maintaining trust at wider distribution scale.
Amazon and Google Disclose Steep Emissions Increases Tied to AI Data Centres
Amazon's 2025 sustainability report revealed a 16% year-on-year rise in greenhouse gas emissions, reaching approximately 81 million metric tonnes of CO₂-equivalent. Google similarly acknowledged that its net-zero trajectory is "non-linear", attributing the deviation primarily to AI data-centre construction. Both companies had previously committed to ambitious carbon targets; these disclosures are the first from either company explicitly linking a significant emissions increase to AI infrastructure growth, adding measurable weight to calls for greener data-centre standards.
Microsoft Forms $2.5 Billion Enterprise AI Company
Microsoft announced the formation of a new $2.5 billion entity, staffed by 6,000 industry and engineering specialists, whose mandate is to design and scale AI systems inside large enterprises. Rather than licensing tools, the company will embed directly with customers to integrate AI into core business processes — a shift toward outcome-based AI services. The move places Microsoft in direct competition with traditional IT consultancies and signals how major AI vendors expect enterprise adoption to mature over the next several years.