Geopolitics of AI – Who Has Access to the Models?

The biggest event of the weekend is a geopolitical earthquake around Anthropic – the US used export controls on AI models for the first time, forcing the company to disable them for the entire world. This triggered a domino effect: the EU protests, US states push their own regulations, and the stock market sees a historic wave of AI giant IPOs. The AI world is no longer just about technology – it is becoming an arena of politics and national security.

Intro · Alex

I have tracked the last 48 hours for you, and I must admit – there hasn't been this much chaos in a long time. First, the shock: the US Department of Commerce forces Anthropic to completely disable Fable 5 and Mythos 5 models for users outside the country. The reason? Amazon reported that its researchers found a way to "jailbreak" the model and extract information about cyberattacks, and concerns arose about access by a Chinese group. This is unprecedented – even allies have been cut off. In this edition, you will find: an analysis of the geopolitical earthquake around Anthropic, Brussels' first reaction, which speaks of "discrimination against partners," an avalanche of state AI regulations in the US despite White House opposition, and a historic wave of IPOs – SpaceX, OpenAI, and Anthropic are going public. Plus, two tech gems that directly affect the security of anyone using AI agents.

What's worth knowing

US Forces Anthropic to Disable Fable 5 and Mythos 5 Models Worldwide

Unprecedented move – the US Department of Commerce issued an export control order, banning access to Fable 5 and Mythos 5 models for all foreigners, including allies. Amazon (CEO Andy Jassy) reported a potential jailbreak and extraction of information about cyberattacks.

EU Commission Reacts to US Decision: "There Cannot Be Discrimination Against Partners"

EU Commission spokesperson Thomas Regnier announced that the EU is "carefully analyzing the practical consequences" and emphasizes that measures should not discriminate against partners. This is Brussels' first official reaction to the US export ban.

US States Enact Own AI Regulations – Despite Trump's Ban

Already 44 states have passed at least one AI-related law, and in 2025 over 100 such acts were adopted. Trump called on Congress to "preempt" state regulations, but states continue to act – regulating chatbots for children and recruitment tools.

From the tech world

High-severity vulnerability in Linux caused by a single faulty character

A single misplaced exclamation mark in the nf_tables subsystem of the Linux kernel (CVE-2026-23111) introduced a use-after-free vulnerability that an unprivileged user can exploit to escalate privileges to root on Debian and Ubuntu.

For the 2nd time in weeks, Microsoft packages laced with credential stealer

Over 70 cryptographically signed open-source packages from Microsoft (including durabletask) were compromised – credential-stealing code was added. The attack was attributed to the TeamPCP group, which used the same stolen OIDC token as in May.

Tip of the day

How to Check if Your AI Agent Has Been Compromised

After over 70 signed Microsoft packages contained credential-stealing code (activating immediately upon opening by an AI agent), it's worth knowing how to protect yourself. Key is verifying package download sources – trusted repositories are not everything. Check if your coding agent (Copilot, Claude Code) has recently downloaded any of the durabletask packages. For devs: enable dependency auditing in CI/CD and check if packages from unofficial channels have appeared in the repository. This is a minimal safeguard, but after this attack – absolutely necessary.