Anthropic puts Claude Mythos 5 into Claude Security in public beta, states partner cyber tools are coming, and adds 35 million dollars in credits for open-source security

Sources: Anthropic blog post, HN discussion
Summary: What shipped is Claude Security, where scans now run on Claude Mythos 5 in public beta for Enterprise customers and are billed as standard token usage. Routing the model's cyber capability through partner defence products is announced and not yet shipped: the post says the model is coming soon to partners' tools, that Anthropic is working with those partners to build it into their products, and that it is early in this work. The stated design is that a user of such a product receives a defined artifact such as a suggested patch rather than a prompt box on the model. The post also commits 35 million dollars in credits directed at open-source security work.
Why it matters: The proposed control is architectural rather than a classifier, so whether it constrains anything depends on partner integrations that do not exist yet.
Follow-up: Track whether the partner cyber-defence integrations ship and what artifact each one exposes.