EU finalizes Digital Omnibus amendments, extending high-risk AI compliance deadline 16 months to December 2027
The EU adopted the Digital Omnibus package on June 30, with the centrepiece AI Act change being a 16-month extension for high-risk AI systems. Stand-alone high-risk AI systems must now comply from December 2, 2027, rather than the original August 2, 2026 date. AI systems embedded as safety components in other products receive further extended timelines. The amendments are designed to allow time for technical standards and compliance infrastructure to be established before enforcement begins.
Why it matters: Companies and nonprofits operating in the EU or shipping products there were racing against an August 2026 deadline that is now gone; the extension materially changes the cost and urgency of AI compliance projects across every sector that uses high-risk AI systems.
Anthropic releases Claude Sonnet 5 as default for all users, closing most of the gap with Opus 4.8 at introductory pricing of $2 per million tokens
Anthropic released Claude Sonnet 5 on June 30, describing it as its most capable Sonnet-class model and the first to substantially narrow the performance gap with Opus 4.8 on agentic benchmarks including BrowseComp and OSWorld-Verified computer use. The model is now the default for Free and Pro Claude.ai plans and is available in Claude Code and via API at introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, before rising to $3 and $15. Safety assessments found Sonnet 5 produces fewer undesirable behaviors than Sonnet 4.6 in agentic contexts.
Why it matters: Near-Opus performance at roughly 60% of Opus pricing collapses the cost-performance tradeoff for developers and organizations running agentic AI workflows, which could accelerate adoption for teams that have been holding back on automation due to cost.
South Korea commits 2,100 trillion won ($1.36 trillion) to AI data centres and semiconductor production as national industrial strategy
South Korea's President Lee Jae Myung unveiled the country's Three Mega Projects initiative on June 29, with SK Group anchoring the AI component by committing 1,000 trillion won to build 15 gigawatts of AI data centre capacity by 2035 (first 5 GW from SK Group, GS Group, and Naver by 2029) and 1,100 trillion won to expand semiconductor production, including accelerating SK hynix's Yongin cluster from a 2045 to a 2033 completion target. SK Group Chairman Chey Tae-won described the goal as transforming South Korea from an "AI-consuming nation" into an "intelligence-exporting nation." The government will support the buildout with region-specific electricity tariffs for AI data centres and a commitment to 100 GW of renewable energy by 2030.
Why it matters: A $1.36 trillion combined investment in AI compute and semiconductor capacity is among the largest single-country AI infrastructure commitments announced anywhere, and it reflects how governments are now treating AI infrastructure as sovereign industrial policy rather than a commercial real estate category.
OpenAI restricts new GPT-5.6 Sol to US-government-approved customers as federal AI model vetting becomes a standard deployment step
OpenAI released GPT-5.6 Sol on June 27 but restricted access to a small group of customers approved by the Trump administration, which has been conducting cybersecurity reviews of frontier AI models before allowing broad deployment. Anthropic faced the same process with Fable 5 and Mythos 5 in mid-June, and by June 29 had received limited approval to roll out Mythos 5 to approved partners. OpenAI said it does not consider the government vetting process "the long-term default" and described restricted access as a temporary step toward broader availability.
Why it matters: Two consecutive frontier model launches being gated through a US government review process establishes a new de facto standard, meaning deployment timelines for powerful new models may no longer be fully under any lab's control.
Anthropic launches Claude Science, a dedicated AI workbench that puts databases, pipelines, and tools in one environment for scientists
Anthropic introduced Claude Science on June 30 at an AI for Science event, describing it as a workbench that gives scientists one environment to conduct computational research across databases, analysis pipelines, and visualization tools without switching between separate systems. Claude Science runs existing Claude models including Opus 4.8 rather than a new biology-specific model, and produces auditable research artifacts. John Jumper, who joined Anthropic from DeepMind earlier in June, spoke at the launch. Claude Science is available now at claude.com/product/claude-science.
Why it matters: Following Claude Code for software development, Claude Science is Anthropic's second vertical environment play, signaling a deliberate strategy to own the operating layer for specific professional domains rather than remaining a general-purpose model API provider.
NSF launches TechAccess AI-Ready America to fund one statewide AI coordination hub per US state, Round 1 proposals due July 16
The National Science Foundation opened solicitation NSF 26-508, which will fund up to 56 state and territorial AI coordination hubs at approximately $1 million per year for three years with a possible fourth-year extension, for a total program budget of $168 to $224 million. Each hub must operate a statewide AI resource navigator, develop a state AI readiness strategic plan, provide hands-on deployment support to small businesses, local governments, and nonprofits, and coordinate sector-specific AI adoption in priority industries. Nonprofits, universities, and state agencies are all eligible as lead applicants; Round 1 will fund 10 hubs, with full proposals due July 16, 2026.
Why it matters: The one-hub-per-state structure means a single designated institution in each state becomes the primary AI adoption infrastructure for that state for at least three years; nonprofits with existing statewide convening relationships have a genuine competitive advantage in this competition over larger universities without that reach.
Google.org commits $10 million to expand AI Opportunity Fund across Asia-Pacific via AVPN, targeting 5.5 million learners
Google.org committed an additional $10 million to its AI Opportunity Fund: Asia-Pacific, channeled through AVPN and the ASEAN Foundation to expand AI literacy across Southeast Asia and the broader region. The expanded fund targets 5.5 million individuals, including 4.7 million learners, through training programs for K-12 teachers, university students and faculty, and micro, small, and medium enterprises. Content draws from Google Skills, Google DeepMind, and Google for Education resources, with the ASEAN Foundation providing in-country delivery infrastructure across member states.
Why it matters: Targeting 5.5 million learners through a coordinated regional platform is the largest AI literacy initiative the Asia-Pacific region has seen, and routing it through AVPN and the ASEAN Foundation gives it delivery infrastructure that individual country programs typically lack.
US Office of the National Coordinator opens $2 million in LEAP grants for agentic AI in clinical care and laboratory data standards
The Office of the National Coordinator for Health Information Technology opened a new LEAP (Leading Edge Acceleration Projects) funding round with $2 million available across three awards: up to $1 million for one project embedding standards-based agentic AI into care delivery and clinical trial settings, and up to $500,000 each for two projects improving laboratory data quality standards and strengthening ONC's Lantern interoperability monitoring tool. All three awards are for a two-year performance period. Health systems, hospitals, nonprofits, universities, and technology organizations are eligible under ONC's standard applicant criteria.
Why it matters: Federal grant funding specifically designed for agentic AI in clinical settings is new; LEAP grants historically shape the direction of health IT interoperability standards for years, meaning this round signals that autonomous AI agents in care delivery are now an ONC infrastructure priority rather than a research curiosity.
Kenya's government holds exploratory talks with Anthropic on applying frontier AI to four national development priorities
Kenya's Special Envoy on Technology Ambassador Philip Thigo held a series of exploratory discussions with Anthropic's Elizabeth Kelly, Head of Beneficial Deployments, focused on applying frontier AI to four specific national priorities: preserving and expanding access to Kenya's 40-plus languages, strengthening pandemic preparedness and disease surveillance, advancing food systems innovation, and improving learning outcomes. No formal partnership was announced, but both sides described the talks as productive with interest in continuing. The engagement follows President Ruto's June meeting with OpenAI CEO Sam Altman on establishing OpenAI Academy in Nairobi.
Why it matters: Kenya naming four concrete development domains rather than a generic AI partnership framing is a more substantive approach than most African governments have publicly articulated, and positions Kenya as the continent's most active bilateral AI diplomacy actor in mid-2026.
AI Impact Hub