Race to AGI Daily Digest - Thursday, June 18, 2026

TLDR

G7 leaders are now directly pressuring the US over the Anthropic export ban, turning a US policy move into a full G7 governance fight.

Read the G7 push on Anthropic’s export ban ->

Zhipu’s GLM‑5.2 has jumped to the top of open‑weight model rankings with a 1M‑token context, expanding what users can pack into a single prompt.

Explore the GLM‑5.2 open‑weight milestone ->

MIT’s DAAAM system gives robots long‑term spatial memory, pushing embodied AI toward real‑world, persistent tasks.

See how DAAAM changes robot memory ->

The EU Parliament is easing parts of the AI Act while banning deepfake undress apps, tightening rules on abuse without freezing research.

Get the latest AI Act changes and deepfake ban ->

Chip‑focused names like ARM and Broadcom rose while Meta, Salesforce, and C3.ai fell, underscoring the premium on infrastructure in a policy‑heavy week.

Scan key AI hardware and software companies ->

The Full Story

Building on Monday’s Fable 5 shutdown, Tuesday’s Pentagon export moves, and Wednesday’s EU sovereignty talk, today the fight jumps to G7 scale. G7 leaders are pressing the US over its Anthropic export ban and what global rules for frontier labs should look like. The message is simple: if models from Anthropic, OpenAI, Google, Cohere, Mistral AI, and Meta are now “strategic assets,” access can’t be decided unilaterally. You can trace that pressure here: G7 push on Anthropic export ban -> and zoom out to the broader access crunch: Anthropic’s Global AI Access Crisis narrative ->. While leaders argue over gates, the frontier keeps moving. Zhipu’s GLM‑5.2 is now topping open‑weight rankings with a million‑token context window. That’s enough to stuff entire codebases, long legal cases, or days of sensor logs into one prompt. If you care about open‑weight options as a counterweight to closed US labs, this is a big one: GLM‑5.2 open‑weight breakthrough -> and Zhipu AI profile ->. Down in the physical world, MIT’s DAAAM system gives robots long‑term spatial memory, letting them remember layouts and tasks across days. That’s a key building block for the embodied robotics pushes we saw from China yesterday. Take a look at how DAAAM works here: MIT’s long‑term memory for robots ->. Europe is finishing its own rulebook in parallel. The Parliament is easing some AI Act rules to keep builders on board, while drawing a hard line on abuse by banning deepfake undress apps—an obvious shot at the darker side of generative image tools. Details are here: EU AI Act tweak and deepfake ban -> and tie straight back to players like xAI ->. Markets are reading this as another day where infrastructure wins. ARM and Broadcom are up, while Meta, Salesforce, and C3.ai slip. That fits our running story: frontier models are being treated as security assets, governments are writing the rules in real time, and the companies selling the shovels for this AI gold rush look increasingly comfortable in the middle.

Want more?

Explore 100 AI companies →Track emerging AI trends →View AI deals graph →

Jun 17, 2026

Jun 19, 2026

Get This Delivered Daily

Join thousands of AI professionals who start their day with Race to AGI.