1. U.S. Triggers Global Export Ban on Anthropic's Fable 5
The White House flagged South Korea's SK Telecom as a China-linked security risk, prompting Anthropic to revoke its access. Amazon researchers then surfaced jailbreak vulnerabilities in the new Fable 5 model, escalating to a worldwide deployment ban on June 12. AI council co-chair David Sacks reportedly gave Anthropic an ultimatum — fix the jailbreaks or de-deploy — and demanded "zero jailbreaks" before any relaunch, a bar security experts call technically impossible for any frontier model today.
2. Anthropic Hits $965B Valuation, Passing OpenAI
Anthropic raised roughly $65 billion in a round valuing the company at $965 billion post-money. That figure surpasses OpenAI's $852 billion private-market valuation for the first time, making Anthropic the most valuable private AI company in the world. The raise lands amid a broader wave of mega-listings, with SpaceX going public on June 12.
3. Claude Fable 5 Posts 95% on SWE-bench Verified
Released June 9, Anthropic's Fable 5 reportedly scores 95% on SWE-bench Verified and 80% on SWE-bench Pro, priced at $10/$50 per million tokens. The coding gains feed Anthropic's claim that its engineers now ship roughly 8x as much code per quarter as in the 2021–2025 baseline. The model's strength is exactly what made its abrupt export ban so disruptive for enterprise users mid-rollout.
4. OpenAI Acquires Astral, Maker of uv and ruff
OpenAI bought Astral, the company behind the widely adopted Python tools uv (package manager) and ruff (linter). OpenAI plans to fold the tooling into its Codex development platform, deepening its grip on the Python developer workflow. The deal followed OpenAI's June 3 "Codex for every role, tool, and workflow" push.
5. Chinese Open-Weight Models Surge: GLM-5.2, Kimi K2.7, MiniMax M3
Zhipu AI shipped GLM-5.2 on June 16, Moonshot AI released Kimi K2.7 Code on June 12, and MiniMax positioned its open-weights M3 frontier model as a ban-proof alternative — stressing that self-hosted models can't be recalled by government directive. Anthropic separately named DeepSeek, Moonshot, and MiniMax as targets of industrial-scale Claude distillation. Qwen3.5-397B-A17B drew attention scoring 88.4 on GPQA Diamond at just $0.60 per million tokens.
6. Enterprise Lock-In: TCS, DXC and Korean Giants Deploy Claude
Tata Consultancy Services will integrate Claude across its 600,000+ employees and clients in banking, healthcare, retail and government, while DXC Technology will pipe Claude into regulated Fortune 500 systems for banks, airlines and hospitals. In Korea, NAVER rolled out Claude Code across engineering, with Samsung SDS, LG CNS, Nexon and Hanwha Solutions announcing deployments touching hundreds of thousands of staff. The deals continued even as the Fable 5 ban played out.
7. Trump Administration Pushes National AI Framework
The White House's National Policy Framework for Artificial Intelligence, released March 20, spans seven pillars and recommends against any new federal AI rulemaking body, favoring existing agencies and industry-led standards. An executive order directs agencies to preempt state AI laws via lawsuits and to weigh conditioning federal grants on state alignment. Senator Marsha Blackburn's proposed "TRUMP AMERICA AI Act" would codify the order into a single national rulebook.
8. Nvidia and Microsoft Unveil RTX Spark Superchip for AI PCs
Nvidia and Microsoft introduced RTX Spark, a superchip integrating GPU and CPU to run AI-agent workloads locally on thin laptops and desktops. Devices from ASUS, Dell, HP, Lenovo and Surface are slated to ship in late 2026. The launch underscores the industry's pivot toward on-device, edge AI agents rather than cloud-only inference.
9. Google Brings Gemini to Search and a New Smart Speaker
Google announced an AI Mode in Search powered by Gemini 3.5 Flash, built to handle multimodal queries directly in results. It also released its first smart speaker in six years, with Gemini integrated for natural conversation and voice interaction. Google additionally shipped DiffusionGemma 26B-A4B, a diffusion-based open model, on June 10.
10. Anthropic Publishes First "Public Record" on AI Cyber Threats
On June 12 Anthropic released its first Public Record, a transparency report documenting AI-enabled cyber threats it detected and mitigated across 2025–2026, including phishing, social engineering and vulnerability analysis. The company also connected Claude to 28 security and compliance platforms — among them CrowdStrike, Palo Alto Networks, Okta and Zscaler — via its Claude Compliance API. Together the moves push Claude deeper into enterprise security stacks.
// KEY TAKEAWAYS
This week geopolitics moved to the center of AI: a U.S. export ban yanked Anthropic's record-setting Fable 5 offline even as the company crossed a $965B valuation and signed massive enterprise deals with TCS, DXC and Korea's biggest firms. The ban handed an opening to China's open-weight camp — GLM-5.2, Kimi K2.7 and MiniMax M3 — whose self-hosted nature makes them immune to government recall. Meanwhile the action keeps shifting from demos to infrastructure: OpenAI buying core Python tooling, Nvidia/Microsoft pushing on-device agents with RTX Spark, and Washington racing to set one national rulebook over state laws.