This Week in AI: June 27 – July 3, 2026
The week's defining thread was resolution. The US government's first serious intervention in frontier AI availability came to a close as export controls on Anthropic's most capable...
On this page+
- The Big Story: US Lifts Export Controls on Fable 5 and Mythos 5 After 19 Days
- Source
- Anthropic Stacks July 1 With Sonnet 5, Claude Science, and a California Statewide Deal
- Source
- OpenAI Previews GPT-5.6: Sol, Terra, and Luna, Under Government-Gated Access
- Source
- Abu Dhabi's MGX Closes $49 Billion AI Fund, One of the Largest Ever
- Source
- Samsung Pledges $648 Billion for AI Infrastructure and Chips in South Korea
- Source
- Qualcomm Acquires Modular for $3.9 Billion to Challenge Nvidia's CUDA Moat
- Source
- Strategic Takeaway
The week's defining thread was resolution. The US government's first serious intervention in frontier AI availability came to a close as export controls on Anthropic's most capable models were lifted, and Anthropic turned the moment into its busiest single product day yet. OpenAI unveiled its next-generation model family under government-gated access, sovereign and corporate capital committed at new scales, and the infrastructure contest behind the AI race reached a new register of ambition.
The Big Story: US Lifts Export Controls on Fable 5 and Mythos 5 After 19 Days
On June 30, the US Department of Commerce lifted export controls on Anthropic's Claude Fable 5 and Mythos 5, ending a 19-day global shutdown triggered on June 12 after Amazon researchers flagged a jailbreak technique that bypassed one of Fable 5's cybersecurity safeguards. Anthropic retrained its safety classifier, reporting the technique is now blocked in more than 99% of cases. Commerce Secretary Howard Lutnick's reversal letter required Anthropic to proactively detect security risks, cooperate with the government on future release protocols, and report malicious activity in its models. Both models returned to all users worldwide on July 1.
Source
Anthropic Stacks July 1 With Sonnet 5, Claude Science, and a California Statewide Deal
Claude Sonnet 5 launched as the most capable Sonnet yet, approaching Opus 4.8 performance on reasoning, tool use, and agentic tasks, at an introductory price of $2 per million input tokens and $10 per million output tokens through August 31, and it becomes the default model for Free and Pro plans. On the same day, Anthropic launched Claude Science, an AI research workbench with pre-configured skills for genomics, proteomics, structural biology, and cheminformatics. Governor Gavin Newsom also signed a deal giving all California state agencies, cities, and counties access to Claude at a 50% discount, the first statewide enterprise AI commitment of its scale in the US.
Source
OpenAI Previews GPT-5.6: Sol, Terra, and Luna, Under Government-Gated Access
On June 26, OpenAI began a limited preview of GPT-5.6, a three-tier model family: Sol for complex research and coding, Terra for high-volume business tasks at competitive performance to GPT-5.5 and 2x lower cost, and Luna for fast everyday work. Access is currently restricted to roughly 20 government-approved companies, with general availability expected in the coming weeks.
Source
Abu Dhabi's MGX Closes $49 Billion AI Fund, One of the Largest Ever
MGX, the two-year-old Abu Dhabi AI investment firm backed by Mubadala and G42, closed its Fund I at $49 billion on July 1, exceeding its $45 billion target. The fund has backed OpenAI, Anthropic, and xAI across their largest raises, drawing capital from investors across the Middle East, North America, Asia, and Europe. Chaired by Abu Dhabi's National Security Adviser, the close confirms Gulf sovereign capital is now a structural fixture of the frontier AI economy.
Source
Samsung Pledges $648 Billion for AI Infrastructure and Chips in South Korea
On June 28, Samsung Group announced a 1,000 trillion won, roughly $648 billion, investment over the next decade covering AI data centers, chip factories, batteries, and displays across South Korea. The commitment, made at a government briefing chaired by President Lee Jae Myung, is the largest corporate investment pledge ever announced by a South Korean company.
Source
Qualcomm Acquires Modular for $3.9 Billion to Challenge Nvidia's CUDA Moat
Qualcomm closed its $3.9 billion acquisition of AI software startup Modular on June 26, pairing its chip hardware with Modular's Mojo programming language and MAX inference engine, software that lets AI models run across Nvidia, AMD, Intel, and Arm hardware without rewrites. The deal gives Qualcomm a credible software layer for its data center push and one of the most direct challenges to Nvidia's CUDA lock-in seen from a major chipmaker.
Source
Strategic Takeaway
The Fable 5 and Mythos 5 restoration closed the US government's first live test of its ability to pull a deployed frontier model offline and negotiate its return. What Anthropic agreed to in exchange sets the new minimum for labs operating at the frontier, and Anthropic used July 1 to demonstrate that cooperation, not just capability, is now part of the competitive equation. OpenAI's government-gated GPT-5.6 preview shows the same pattern taking hold across both leading labs. Meanwhile, MGX at $49 billion, Samsung at $648 billion, and Qualcomm's Modular deal all point to the same underlying contest: who controls the full stack from silicon to software. The clearest signal of the week is that the race is no longer just about which lab ships the most capable model. It is about who has secured the infrastructure, the capital, and the government relationships to operate at the frontier long-term.
