Anthropic will give the EU early access to its powerful Mythos AI model, allowing cybersecurity experts to test potential risks before a wider rollout.