If the past few weeks were about AI's money-and-power phase, this week was about who actually holds the power. Washington kept one of Anthropic's most capable models switched off worldwide, America's energy regulator moved to rewire the grid around AI data centres, and the field's most sought-after engineer walked from Google to OpenAI — all while Elon Musk's empire spent US$60 billion to bolt a coding company onto xAI. The throughline is hard to miss: governments and a handful of giants are now shaping the AI industry as much as the labs themselves.

Washington keeps Anthropic's Fable 5 switched off — worldwide

On 12 June, Anthropic received a US government export-control directive ordering it to suspend access to its newest models, Claude Fable 5 and Claude Mythos 5, for any foreign national — whether inside the US or not. Because the company couldn't reliably verify every user's nationality in real time, it couldn't comply selectively, so it switched both models off for everyone, Americans included. More than a week later they remain dark. The government's stated concern was a newly discovered method of "jailbreaking" Fable 5, though its letter reportedly gave no specifics. The White House has since told Anthropic it must identify and eliminate all jailbreaks before relaunch — a bar security researchers near-unanimously say no frontier model can clear today. You can read Anthropic's own statement and Fortune's reporting for the detail.

Why it matters: This is the starkest reminder yet that a model your product depends on can vanish overnight for reasons that have nothing to do with you. It is a powerful argument for multi-provider design rather than hard-wiring a single vendor — which is exactly how our team approaches building and deploying AI for clients. When the model layer can be switched off by regulators, resilience stops being a nice-to-have and becomes basic engineering hygiene.

Google loses its transformer architect to OpenAI

Noam Shazeer — co-author of the 2017 paper "Attention Is All You Need", which introduced the transformer architecture underpinning virtually every major AI model today, and co-lead of Google's Gemini family — announced on 18 June that he is joining OpenAI as its Lead for AI Architecture Research. Google had paid roughly US$2.7 billion in 2024 to bring Shazeer and his team back from Character.AI, making his departure an unusually expensive one to absorb. Sam Altman called the hire one "10 years in the making". Silicon Republic has the background.

Why it matters: In a year when frontier models are separated by weeks rather than quarters, talent is the scarcest resource of all. Losing the person most associated with the transformer is a symbolic blow to Google just as it prepares to ship Gemini 3.5 Pro, and it signals OpenAI doubling down on fundamental model architecture in the run-up to its expected public listing. The poaching war between the labs is now as consequential as the model race itself.

America's energy regulator moves to rewire the grid for AI

On 18 June, the US Federal Energy Regulatory Commission issued tailored "show cause" orders to six regional grid operators, directing each to either defend its existing rules or propose reforms that let large-load customers — explicitly AI data centres — connect to the grid faster while protecting reliability and consumer costs. FERC Chair Laura Swett called AI grid integration a "national priority", and the orders deliberately bypass the years-long normal rulemaking process in favour of something that can move in weeks. The American Action Forum has a clear explainer.

Why it matters: Power, not chips, is becoming the real bottleneck for AI. Hyperscalers are adding gigawatts of capacity faster than grids can absorb them, and the fight over who pays for those upgrades is turning into one of the most politically charged stories of the entire build-out. It is a debate Australia knows well, with our own data-centre boom raising near-identical questions about energy and water — expect the regulatory playbook being written in Washington to echo here.

SpaceX buys Cursor for US$60 billion to arm xAI

In an SEC filing around 16 June, SpaceX disclosed a US$60 billion all-stock acquisition of Anysphere, the company behind the popular AI coding tool Cursor. The deal folds Cursor into Elon Musk's broader xAI ecosystem: the two have been jointly training a coding model for months, which is expected to ship inside both Cursor and xAI's "Grok Build" agent, with SpaceX supplying GPU cluster capacity. Cursor reportedly generates around US$4 billion in annualised revenue, and the deal is expected to close in the third quarter pending regulatory approval. CBS News covered the filing.

Why it matters: AI coding is the most lucrative battleground in the industry, and Musk is buying his way into a fight currently led by Anthropic's Claude Code and OpenAI's Codex. Coming hot on the heels of SpaceX's own blockbuster IPO — which followed the OpenAI listing plans we covered last week — it is another sign that compute and distribution now matter as much as the models themselves.

Anthropic plants a flag in Asia-Pacific

Even amid the Fable 5 fallout, Anthropic pressed ahead with its regional expansion, opening a Seoul office on 17–18 June — its third in Asia-Pacific after Tokyo and Bengaluru. The launch came with a wave of Korean enterprise deployments: NAVER has rolled out Claude Code across its entire engineering organisation, while Samsung SDS, LG CNS, Nexon and Hanwha are all adopting Claude in various forms. The company also signed a memorandum of understanding with Korea's science ministry on AI safety and model evaluation. Anthropic says weekly Claude Code users in Korea have grown sixfold in four months. The full list is on Anthropic's newsroom.

Why it matters: The enterprise land-grab is global and accelerating fastest in our part of the world. For Australian businesses, the speed at which regional neighbours are standardising on these tools is a useful signal: the question is shifting from "should we adopt AI" to "how do we deploy it well", even as the Fable 5 episode shows why governance and contingency planning have to come along for the ride.

AI chatbots are quietly becoming a news front door

The Reuters Institute's Digital News Report 2026, published this week, found that weekly use of AI chatbots for news has climbed to 10% of audiences globally, up from 7% a year ago — and among the under-35s, it is closer to one in six. The catch for publishers: only around 42% of those users say they always or often click through to the original source, and trust in AI-chatbot news still sits at roughly 20%, well below traditional outlets.

Why it matters: This one lands close to home for anyone who publishes online. If audiences increasingly get their answers summarised inside a chatbot without ever visiting the source, referral traffic and the ad models built on it come under real pressure. The flip side is that being accurately cited by an AI assistant is becoming its own discovery channel — which makes well-structured, quotable content more valuable, not less.

Quick Hits

  • Grok lands on AWS. xAI's Grok 4.3 is now generally available on Amazon Bedrock, with a one-million-token context window and aggressive pricing, giving enterprise AWS teams direct access without a separate xAI contract (AWS).
  • OpenAI courts the consultants. OpenAI launched its first formal Partner Network on 14 June, backed by US$150 million and aiming for 300,000 certified consultants by year-end, with Accenture, BCG and McKinsey among the founding partners.
  • Gemini 3.5 Pro still waiting. Google's next flagship remains in limited Vertex AI preview as its self-imposed June general-availability window narrows; confirmed specs include a two-million-token context window and a "Deep Think" reasoning mode (Vertex AI release notes).
  • Amazon shelves the Altman biopic. Amazon dropped Luca Guadagnino's nearly finished Sam Altman drama "Artificial" — an awkward move given Amazon's reported US$50 billion investment in OpenAI (Variety).
  • Europe's clock keeps ticking. The European Commission published a new Code of Practice on labelling AI-generated content as the EU AI Act counts down to 2 August, when most of its core obligations become enforceable.

Looking Ahead

The story to watch is whether Anthropic and Washington can agree a workable monitoring framework that brings Fable 5 back online — and what precedent it sets if they can't. Keep an eye, too, on whether Google ships Gemini 3.5 Pro before the month is out. If you want to keep track of the tools behind all this churn, we maintain a running library of AI tools and resources worth knowing.

That's the wrap for this week. The pace isn't slowing, so check back next week for the next instalment of What's New in AI — we'll keep cutting through the noise so you don't have to.