AI news5
US government issues first-ever export control directive against an AI company, forcing Anthropic to suspend Fable 5 and Mythos 5 globally
On June 12, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei ordering the company to suspend access to its Fable 5 and Mythos 5 models for "any foreign national" inside or outside the US, citing risk of diversion to foreign military or intelligence use. Anthropic complied by disabling access globally, affecting its own international employees. The Trump administration has not publicly disclosed the legal basis for the order. More than 80 cybersecurity leaders including Nvidia and Adobe signed an open letter protesting the action; it is the first time the US has exercised export controls against an AI company.
computerweekly.com ↗Why it matters
a government directive that grounds a frontier AI model globally — with no public explanation, no legal precedent, and collateral impact on the company's own workforce — sets a template for state control over AI deployment that no other country has yet attempted at this scale, and is prompting AI sovereignty debates from India to Europe.
Nature publishes two medical AI systems — MIRA and Google's AMIE — that match physician performance across the full patient management cycle
Two independent studies in Nature this week present MIRA (Medical Intelligence for Reasoning and Action) and Google's AMIE (Articulate Medical Intelligence Explorer), both capable of managing multiple stages of patient care from diagnosis through treatment decisions. Unlike earlier medical AI systems that specialise in narrow tasks, both models handle patient histories, investigations, diagnosis, treatment planning, and outcome monitoring. In independent evaluations, both performed at least as well as physicians on clinical management benchmarks.
medicalxpress.com ↗Why it matters
matching physician performance across the full care cycle — not just diagnosis — is the threshold that makes AI clinically useful in settings with physician shortages; the Nature peer-review bar also raises the evidentiary standard well above proprietary benchmark claims.
World model startup Odyssey raises $310M Series B at a $1.45B valuation, backed by Amazon, AMD Ventures, and GV
Odyssey, founded by autonomous vehicle veterans Oliver Cameron (ex-Voyage/GM Cruise) and Jeff Hawke (ex-Wayve), has closed a $310 million Series B led by Natural Capital at a $1.45 billion valuation. The company builds world models — AI systems trained on physical environments that simulate the world with accurate physics — for use cases including video game creation, robotics, and generating rich interactive video from text prompts. Odyssey gathered training data by sending people out with body cameras, analogous to how Google Maps captured street-level imagery. AWS is now its preferred cloud provider.
techcrunch.com ↗Why it matters
world models represent the next wave of AI beyond language and image generation — systems that reason about how the physical world works; a $1.45B valuation for a company founded in 2023 signals that the market believes physics-grounded simulation is foundational infrastructure for robotics and interactive media.
Vercel launches "eve," an open-source agent framework that ships production infrastructure with every agent by default
Vercel released eve (Apache 2.0, github.com/vercel/eve) on June 17, an open-source framework for building, running, and scaling AI agents in which each agent is structured as a directory of files defining its model, instructions, tools, skills, subagents, and schedules. Eve bundles the infrastructure that builders currently hand-roll themselves: durable execution, sandboxed compute, human-in-the-loop approvals, subagents, and evals. Vercel positions it as the Next.js for agents — eliminating the plumbing problem that currently means every agent team rebuilds the same stack from scratch.
vercel.com ↗Why it matters
if eve achieves what Next.js did for web development, it shifts agent development from a systems engineering project to a product engineering project — lowering the barrier enough that smaller teams and nonprofits can build and deploy reliable agents without dedicated infrastructure expertise.
OpenAI launches scheduled tasks in ChatGPT, replacing Pulse with a persistent task scheduler for Plus, Pro, Business, and Enterprise users
OpenAI rolled out scheduled tasks across ChatGPT on June 17 for paid users, replacing Pulse — the existing proactive tasks feature — which will be removed in 14 days. The new system adds a dedicated Scheduled sidebar where users can view, pause, resume, edit, or delete active tasks; schedule work for specific times or broader windows (morning/afternoon/evening); and set monitoring tasks that search the web and connected apps for changes, notifying users only when something worth reporting happens. Tasks are capped at once per hour and may auto-pause after inactivity.
9to5mac.com ↗Why it matters
persistent, user-defined task scheduling in a consumer AI app is the first mainstream rollout of the "AI that acts on your behalf over time" model — distinct from single-session assistants; its reach across Plus/Pro/Business/Enterprise makes it the widest deployment of agentic scheduling to date.
AI in the nonprofit sector4
Code.org rebrands to CodeAI and the EU Commission and OECD jointly release an AI Literacy Framework for K-12 schools
Code.org, one of the largest K-12 computer science curriculum providers globally, announced it is rebranding to CodeAI and expanding its mission from coding education to AI literacy and "digital fluency." The rebrand comes as the European Commission and OECD simultaneously published an AI Literacy Framework for primary and secondary education on June 18, developed with CodeAI's support, organising student competences across four domains: engage with, create with, manage, and shape AI. Code.org CEO Karim Meghji said the goal is to teach students how AI works under the hood — including model training, data, and hardware — not just how to use AI tools.
edweek.org ↗Why it matters
two of the largest players in K-12 education policy moving in the same direction at the same time — a major US nonprofit redefining its mission and the EU/OECD releasing a shared framework — signals that AI literacy is crossing from a pilot topic to a foundational curriculum expectation, which will shape how a generation understands and engages with AI.
Anthropic becomes the first AI startup to join the Frontier carbon removal coalition, contributing to a new $915M tranche that brings total pledges to $1.8B
Frontier — the advance market commitment for carbon removal founded in 2022 by Stripe and Google — announced a new $915 million funding tranche on June 17, with Anthropic joining as the first pure AI company. The coalition, which also includes Shopify, Salesforce, and H&M Group, commits to buying future carbon removal credits to derisk projects and pull them toward commercial scale. Frontier has contracted nearly $700 million across 50+ projects, delivering approximately 23,000 tons in 2025. The new tranche will focus on 10-15 concentrated bets with 8-10 year offtake contracts extending to 2040.
techcrunch.com ↗Why it matters
Anthropic's arrival as the first AI startup in Frontier is notable against the backdrop of AI companies being among the largest new electricity consumers on the planet; it also signals that the largest AI companies are beginning to treat carbon commitments as a reputational necessity on par with safety disclosures.
Resilience AI deploys hyperlocal climate disaster management plans across 75 gram panchayats in India, funded by the UN and UNICEF
Resilience AI, an enterprise tech and climate resilience startup funded by the United Nations and UNICEF, has deployed AI-powered disaster management plans across 75 gram panchayats (village-level government units) in Uttar Pradesh, Bihar, and Tamil Nadu. The system uses hyperlocal risk data — combining satellite imagery, historical disaster records, and community-level inputs — to produce actionable preparedness and response plans tailored to each village's specific climate risks. The initiative targets communities in flood, cyclone, and heat-wave-prone areas with limited access to centrally-produced risk assessments.
yourstory.com ↗Why it matters
most climate AI tools target national or regional governments; hyperlocal deployment at the village unit — where the response gap is largest — with UN and UNICEF funding behind it is one of the clearest examples to date of AI being used to close the last-mile gap in climate preparedness for the communities most exposed to harm.
Irish nonprofit Camera Education secures €2.5M from UNICEF to expand AI-powered digital education across Ethiopia
Camera Education, an Irish nonprofit, has secured more than €2.5 million from UNICEF to scale its AI-powered digital education program across Ethiopia, where large numbers of students have no prior access to technology. The program teaches digital skills including computing fundamentals and AI literacy through a structured curriculum delivered via low-cost devices, with a focus on students who have never interacted with a computer or smartphone. The UNICEF funding supports expansion to new regions and the adaptation of curriculum materials for local languages and contexts.
siliconrepublic.com ↗Why it matters
digital education programs funded at this scale by UNICEF typically signal readiness for national-level integration — a €2.5M grant in this context is meaningful infrastructure funding, not a pilot, and positions AI-literacy-from-scratch as a viable model for low-infrastructure education systems.