This Week in AI: June 15-21, 2026 (Week 25)

News of the week: June 15-21, 2026 (Week 25)

The week was dominated by Anthropic’s dispute with the US government, which kept its most powerful models, Fable 5 and Mythos 5, switched off. The market saw several big deals: SpaceX is buying Cursor for $60 billion and Salesforce is acquiring Fin for $3.6 billion. The battle for talent between the labs continued, with Noam Shazeer leaving Google DeepMind for OpenAI and Nobel laureate John Jumper leaving for Anthropic. The European Parliament approved a delay to parts of the AI Act, and the US regulator FERC sped up grid connections for data centers.

Anthropic and the US Government

Fable 5 and Mythos 5 Remain Offline

In mid-June, Anthropic switched off its two most powerful models, Claude Fable 5 and Claude Mythos 5, for all customers. The reason was a letter from the US Department of Commerce dated June 12 which, citing export controls, barred access to the models for anyone without US citizenship, including the company’s own employees.

According to TechCrunch (June 15), the letter gave no specific justification and was not made public. Anthropic believed it was related to a discovered way of bypassing its safety guardrails (a so-called jailbreak). Security researchers including Katie Moussouris, however, objected that such a flaw should never trigger an export control. (TechCrunch)

On June 15, veterans of the cybersecurity field publicly called the ban “dangerous” because it takes one of the most capable tools away from defenders. (TechCrunch)
On June 17, CNBC pointed out that Anthropic had long called for regulation itself, but Washington went much further; the company has not yet said when the models will return. (CNBC)
On June 17, TechCrunch reported that world leaders want American AI, but they do not want the US to be able to “turn it off” at any time. (TechCrunch)

So Far the Dispute Seems to Be Helping the Company

Data from Ramp covering more than 70,000 businesses show that Anthropic’s share of business AI subscriptions rose to 41% in May, while OpenAI had 39.5%. This makes it the first time Anthropic has overtaken its rival in business spending (June 16).

At the end of May, the company raised $65 billion at a valuation of $965 billion. (TechCrunch)

Expansion into Korea and a Carbon Coalition

On June 17, Anthropic opened an office in Seoul, its third in the Asia-Pacific region after Tokyo and Bengaluru. The office is led by KiYoung Choi, and the company announced new partnerships within the Korean AI ecosystem. (Anthropic)

On the same day, Anthropic became the first pure AI company to join the Frontier coalition, which makes advance purchases of carbon removal from the atmosphere. The coalition secured new commitments worth $915 million, bringing the total to $1.8 billion. (TechCrunch)

Business and Investment

SpaceX Is Buying Cursor for $60 Billion

On June 16, SpaceX announced it will buy Cursor, the popular AI-assisted coding tool, for $60 billion in its own stock. The deal came just days after SpaceX’s record-breaking stock market debut (IPO) and is expected to close in the third quarter of 2026.

The companies had already formed a partnership in April, when SpaceX secured the right to either buy Cursor or pay $10 billion. Before this offer, Cursor had been in talks about an investment at a $50 billion valuation. (TechCrunch)

Salesforce Acquires Fin

On June 15, Salesforce announced the acquisition of the Fin platform (formerly Intercom) for roughly $3.6 billion. Fin builds AI agents for customer support, meaning programs that independently answer and resolve customer requests.

Fin’s team and technology are set to strengthen the Agentforce platform, with completion expected in the fourth quarter of fiscal year 2027. (TechCrunch)

Amazon Wants to Sell Its AI Chips

According to the head of AI at AWS, Peter DeSantis, Amazon is in talks to sell its Trainium chips to other companies for their own data centers (June 18). This would mean more direct competition for Nvidia; according to the company, Trainium capacity is currently sold out. (TechCrunch)

Other Investments and Deals

On June 15, the Indian company Sarvam became a “unicorn” (a startup valued at over $1 billion): it raised $234 million at a $1.5 billion valuation, in a round led by HCLTech. (TechCrunch)
On June 17, world model (AI that simulates an environment and its physics) maker Odyssey raised $310 million at a $1.45 billion valuation; Amazon also invested. (TechCrunch)
On June 18, according to the WSJ, Baseten, a company that runs AI models (inference), is raising $1.5 billion at a $13 billion valuation. (TechCrunch)
On June 18, Snap spun off its AI video team into a new company, Dotmo, due to costs; it will develop models for interactive games. (TechCrunch)

People and Talent

Noam Shazeer Leaves for OpenAI

On June 18, Noam Shazeer, co-author of the Transformer architecture (the foundation of today’s language models) and co-lead of Gemini development, announced he is leaving Google DeepMind for OpenAI. Google had brought him back in 2024 as part of a $2.7 billion deal with Character.AI.

OpenAI also hired Dean Ball, a former White House AI adviser, who will lead the new Strategic Futures team starting July 6. Both hires fit into OpenAI’s preparations for its stock market debut. (TechCrunch)

John Jumper Heads to Anthropic

On June 19, John Jumper, winner of the 2024 Nobel Prize in Chemistry and leader of the AlphaFold team (AI for predicting protein structures), announced that after almost nine years he is leaving DeepMind to join Anthropic. Google thus lost two prominent figures within a single week. (TechCrunch)

Products and Models

Android 17 and a New Google Home Speaker

On June 16, Google released Android 17 and Wear OS 7. An accompanying update for Pixel phones added the Lyria 3 music model, conversational video editing through Gemini Omni and speech translation. (TechCrunch)

On June 17, Google introduced the Google Home Speaker, a $99.99 smart speaker built around the Gemini assistant. It is the company’s first standalone speaker since 2020, and advanced features require a Google Home Premium subscription. (TechCrunch)

Alibaba Enters Robotics

On June 16, Alibaba unveiled its first suite of AI models for robots, the Qwen Robot Suite. It consists of three models: one for navigation, one for object manipulation and a so-called world model that predicts how a scene will change.

The suite is being piloted with business customers on the Unitree Go2 four-legged robot. (SCMP)

Smaller News

On June 15, Meta brought to Facebook a search feature called “AI Mode“, which builds answers from public posts, groups and videos using Meta AI. (TechCrunch)
On June 17, ChatGPT added a Scheduled page with an overview of scheduled tasks that can be paused, edited or deleted. (OpenAI)
On June 17, the Gemini API added support for streaming generated speech with the gemini-3.1-flash-tts-preview model; on June 15, Google announced the retirement of the older Imagen 4 and Veo 2.0/3.0 models. (Google AI for Developers)
On June 18, Adobe added its AI assistant Firefly to Premiere, Illustrator and InDesign, where it can handle tasks such as batch renaming clips. (TechCrunch)
On June 21, TechCrunch summarized the practical Apple Intelligence features in the iOS 27 beta, such as reading receipts to split payments or automatically replacing weak passwords. (TechCrunch)

Market and Public Opinion

According to a Sensor Tower report, ChatGPT’s share of the AI assistant market fell below half for the first time at the end of May, to 46.4%. Gemini had 27.7% and Claude 10.3% (published June 16). (TechCrunch)

Public attitudes remain cautious. According to a study published on June 17, only 16% of Americans think AI will have a positive impact on society. (TechCrunch)

Regulation and Law

European Parliament Approves AI Act Changes

On June 16, the European Parliament approved the so-called Digital Omnibus on AI, a package simplifying the European law on artificial intelligence (the AI Act). 423 MEPs voted in favor, 57 against and 174 abstained.

Among other things, the amendment postpones obligations for high-risk AI systems to 2027 and 2028 and bans apps that create fake nude images from photos. The final step is formal approval by the Council of the EU. (iubenda, Council of the EU)

Faster Grid Connections for Data Centers

On June 18, the US Federal Energy Regulatory Commission (FERC) unanimously ordered six major grid operators to speed up data center connection requests. The operators have 30 days to report available capacity and 60 days to defend or revise their rules. (TechCrunch)

xAI Turbines as a Matter of National Security

On June 15, the US Department of Justice sided with xAI in a lawsuit brought by the NAACP over unpermitted gas turbines at data centers in Memphis. The department called them a matter of national, economic and energy security; xAI already operates 57 of them. (TechCrunch)