Daily Briefing — June 17, 2026

Over the weekend, Anthropic announced it will "abruptly disable" its most advanced AI models, Fable 5 and Mythos 5, for all users after the US government issued an export control directive ordering the company to cut off access for foreign nationals. The government didn't give Anthropic the full picture on why. What Anthropic does know is that officials believe someone figured out how to jailbreak Fable 5 in a way that lets it be weaponised for identifying software vulnerabilities.

Anthropic had been in a rough patch with the Trump administration over its IPO ambitions, and things were apparently just starting to thaw. Then this lands. Whether this is a real security emergency or partly a negotiating move in a longer political chess game, nobody outside the administration seems to know.

Frontier AI models are now firmly inside the national security conversation. Governments are starting to treat access to powerful AI the same way they treat access to advanced semiconductors or weapons-grade software. The "open access everywhere" era for top-tier models is closing, and more stories like this are coming.

Back in April, SpaceX announced it had an option to acquire Cursor for $60 billion, with a $10 billion breakup fee if it walked away. It's not walking away. SpaceX confirmed the full acquisition as an all-stock deal expected to close in Q3, two days after its IPO and months after merging with xAI. The option became a commitment.

Cursor was early. It was one of the first tools to bake LLM features into a development environment rather than bolt them on as an afterthought. But early does not mean dominant forever. Claude Code has eaten significant market share, Cursor was reportedly struggling to break even, and its leadership openly admitted that compute access was the ceiling on its growth. This looks more like a lifeline dressed up in a press release than a triumphant exit.

The strategic logic from xAI's side is pretty clear. They already struck a compute access deal with Cursor earlier this year and started co-training models together, including Grok Build. Buying Cursor outright gives xAI a direct distribution channel into developer workflows. If you can own the tool that sits between a developer and their codebase, you own a lot of mindshare.

The $60 billion number deserves some skepticism though. That is a significant valuation for a product that was losing ground and bleeding toward breakeven. All-stock deals at elevated valuations right after a fresh IPO tend to be more generous on paper than in practice.

Pokémon Go data might be helping military drones navigate war zones. Niantic, the company behind the game, collected location scan data from players who opted into a 2021 feature that rewarded them for scanning real-world environments with their phones. Before Niantic sold its gaming division in 2025, that data was used to train an AI model capable of recognising and interpreting physical spaces. That model is now potentially being used to help military drones find their bearings in conflict zones.

The bigger thing to sit with is the gap between what people thought they were doing (catching virtual creatures for fun) and what their data ended up doing (training AI for battlefield navigation). Eight hundred million downloads worth of casual players scanning their surroundings had no reasonable way to anticipate this outcome. Most of them probably didn't read the opt-in terms closely enough to see it coming either.

When you contribute data to a consumer product, you're often seeding capabilities that will outlive the product and travel far outside the context you imagined. The gaming division got sold. The data stayed useful. That gap between intention and outcome keeps widening, and it's relevant well beyond the defense sector.