Welcome back to AI Weekly Updates. Each Monday we highlight the most important stories from the past week so you can stay ahead without wading through endless headlines. The week belonged to Anthropic — a government order that pulled its top models offline, a sweeping coding study, a $150M fellowship, and a new Seoul office — set against a busy funding tape running from DeepSeek to Bengaluru.
U.S. export order forces Anthropic to suspend Fable 5 and Mythos 5
The Trump administration directed Anthropic to block foreign nationals — including its own overseas employees — from its newest models, and the company responded by pulling Fable 5 and Mythos 5 offline entirely, even for U.S. users. A letter signed by 76 security leaders, among them CISOs, founders, and researchers, told Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross the move was dangerous, arguing it strips the strongest tools from defenders while rivals keep advancing. The directive reportedly followed Amazon researchers slipping past some of Fable 5’s anti-hacking guardrails. We unpacked what the Fable 5 and Mythos-class permission system actually is when the model launched.
Read more ↗DeepSeek lands a reported $7.4 billion in its first outside round
China’s DeepSeek took its first external financing — roughly $7.4 billion, the largest startup deal anywhere outside the U.S. last week — but on unusually lopsided terms. Per reporting cited by Crunchbase, backers bought into an LLC controlled by founder Liang Wenfeng rather than the company itself, accepting a five-year lockup and no voting rights. It is a striking sign of how far investors will bend to get exposure to a frontier Chinese lab.
Read more ↗Anthropic study finds domain expertise, not coding skill, drives agent success
Anthropic analyzed roughly 400,000 Claude Code sessions from about 235,000 people between October 2025 and April 2026 and found a consistent split: users make around 70% of the planning calls while Claude handles about 80% of execution. What separates a successful session is the user’s grasp of the problem — not whether they code for a living — with every major occupation finishing near the rate of software engineers. Over the seven months, the share of work spent debugging roughly halved and the estimated value of a typical task rose about 25%.
Read more ↗Odyssey raises $310 million for AI world models, the week’s largest U.S. round
Menlo Park-based Odyssey closed a $310 million Series B at a $1.45 billion valuation, led by Natural Capital with Amazon, AMD Ventures, EQT, Google Ventures, IQT, and SignalRank joining. The company builds AI “world models” that generate interactive, multi-sensory simulations of physical environments, and has now raised about $337 million to date. It was the biggest U.S. venture round of the week in an otherwise quiet stretch for mega-deals.
Read more ↗Anthropic puts $150 million into Claude Corps, placing 1,000 fellows in nonprofits
Anthropic launched Claude Corps on June 18, a $150 million program that trains 1,000 fellows and pays them $85,000 to spend a year full-time and in-person helping as many as 400 U.S. nonprofits adopt AI. CodePath serves as employer of record and Social Finance will evaluate the program and build a vehicle to scale it; applications are open to anyone over 18 with under two years of experience, regardless of degree. The first cohort of 100 starts in October 2026.
Read more ↗Genspark reaches a $2.6 billion valuation on a fresh $100 million round
Genspark.ai said on June 17 it had raised $100 million in an extended round that values the Palo Alto company at $2.6 billion, backed by existing investors including Sozo Ventures, Korea Mirae Asset, and UpHonest Capital. The raise brings its total Series B to $485 million for an AI workspace that strings together multiple models to build presentations, financial models, and apps. The company says more than 6,000 businesses signed on within six months and that it added $150 million in annual recurring revenue in the first quarter alone.
Read more ↗Anthropic opens a Seoul office as Naver and Samsung scale up Claude
Anthropic opened its Seoul office on June 17 and signed a memorandum with Korea’s Ministry of Science and ICT to collaborate on AI safety and cyber threats. Naver has now rolled out Claude Code across its entire engineering organization, while LG CNS, Samsung SDS, and Hanwha are deploying Claude to thousands of employees. Korea ranks among the top dozen countries for Claude usage, and the new office is hiring.
Read more ↗AI inference platform Baseten is reportedly raising $1.5 billion
Baseten is reportedly raising a $1.5 billion round, with Spark Capital, Sands Capital, Altimeter Capital, and Wellington Management named among the investors, though the deal had not been finalized as of June 18. The company runs an inference platform that routes requests across AI models to tune for speed and cost — and the raise would land roughly five months after a $300 million Series E in January. The appetite underscores how much capital is flowing into the plumbing beneath AI applications.
Read more ↗Sarvam raises $234 million to become India’s newest AI unicorn
Sarvam AI announced the first close of a $234 million Series B on June 15, with HCLTech as strategic lead and Bessemer Venture Partners, Khosla Ventures, and Peak XV Partners participating. The round vaults the company to unicorn status as it positions itself as a full-stack player spanning its own model development and enterprise applications. It is a notable marker for India’s ambition to build frontier AI at home rather than import it.
Read more ↗Gradial raises $65 million to bring AI agents to marketing operations
Gradial closed a $65 million Series C led by Insight Partners for an agentic marketing platform that coordinates work across the tools teams already run, with an eye toward both efficiency and compliance. The company plans to expand its roughly 100-person team and deepen its Salesforce and Adobe integrations. For marketing leaders, it is a concrete read on where agent budgets are heading: into the orchestration layer that sits on top of the existing stack.
Read more ↗
That wraps this week’s roundup. The headline tension — a frontier lab forced to throttle its best models while investors race to fund the next ones — is the AI story of 2026 in miniature. Check back next Monday for another edition of AI Weekly Updates.