AI access becomes diplomatic currency

TL;DR: The European Union just secured access to OpenAI’s new cyber model, CNBC reported today—while Anthropic’s more powerful Mythos remains off-limits to most governments. Meanwhile, President Donald Trump arrives in Beijing this week to discuss AI with Chinese President Xi Jinping officially for the first time. Powerful AI models have quickly become a strategic asset, with governments and companies racing to lock in access before rivals do.

What happened: The EU Commission welcomed OpenAI’s decision to grant European partners access to GPT-5.5-Cyber, a version of the AI lab’s latest model only available for vetted cybersecurity defenders. George Osborne, head of the OpenAI for Countries initiative, said that “the latest cyber AI capabilities should be available for Europe’s many defenders, not just the few.” Though the offer is likely just as much about competitive positioning against Anthropic as it is a diplomatic gesture—a Commission spokesperson told CNBC that similar discussions with Anthropic were “not yet at the same stage.”

They’re not the only talks that are happening. After the US accused China of using “industrial-scale” distillation to “steal American AI” just a few weeks ago, the question of working together on AI safety is now on the Trump-Xi agenda—with tech titans like Elon Musk and Tim Cook joining Trump in Beijing this week.

How the EU got here: While Anthropic has been in talks with the EU about testing European banks for Mythos-found security flaws, it skipped a European Parliament hearing on its cyber risks last week, which one Dutch lawmaker called “extremely worrying.” The UK's AI Security Institute got hands-on evaluation access to Mythos, but it seems no EU government or company has.

Last month, Germany’s central bank warned that European governments need to request access to Mythos to shield their financial institutions from cyberattacks—something several American banks are already in the process of, potentially giving them an advantage. Hackers are already using Mythos-like AI to create zero-day exploits that can break into widely used software, as Google reported today.

The China-US AI pivot: This is the first time the US has engaged in diplomatic talks with China on AI since former President Joe Biden in 2024. Xi and China’s foreign minister had privately noted that AI’s power would eventually require cooperation with the US, former national security adviser Jake Sullivan told the LA Times—but Mythos appears to have sped up that timeline.

Bottom line: By August, the EU will be able to use its AI Act to demand access to systemic-risk models—if labs refuse, they face fines of up to 3% of global revenue or 15 million euros ($17.7 million). But there’s no guarantee they’ll comply, and access to frontier AI that could have security implications for foreign governments and companies could still be dictated in Washington, Beijing, or just a few San Francisco offices. —WK

Also at OpenAI…

• It just announced its new AI consulting firm, which it says will help other companies build and deploy their own AI systems.