Dimon warns Claude Mythos could become a cyber weapon

JPMorgan CEO Jamie Dimon says Anthropic’s Claude Mythos raises serious security concerns, comparing unrestricted access to the model to “giving ballistic missiles to individuals.” He made the remarks at the Pennsylvania Defense and Innovation Summit.

The US government has already instructed Anthropic to block access to Mythos for foreign nationals, citing security concerns. Public access is also restricted, with the model currently limited to a small group of organizations, companies, and federal and military departments.

“You’re giving ballistic missiles to individuals.”

Claude Mythos cybersecurity capabilities

Mythos is not a self-aware system capable of taking over military equipment like Skynet in The Terminator. Its danger comes instead from its ability to identify cybersecurity vulnerabilities, report them, and automatically generate working exploits.

That makes the model a powerful defensive tool for white-hat security analysts. In the wrong hands, however, the same capabilities could help attackers scale ransomware, phishing, and other data-based scams, making cybercrime more profitable.

Recommended reading

Gemini 3.5 Pro Misses Google’s June Launch Promise

The risks extend beyond cybersecurity. Attackers could potentially use the model to process complex scientific concepts, including biological information, in ways that could enable misuse.

Dimon also highlighted the strategic imbalance that could arise if individual nations or organizations gained exclusive access to the technology.

Will Claude Mythos become publicly available?

Given the security concerns, Claude Mythos is unlikely to become publicly available soon. Anthropic has said it intends to make Mythos-level features and capabilities more widely accessible, but only after developing and applying suitable safeguards.

One possible approach would be to offer public access while restricting features related to cybersecurity and biological or medical research. Project Glasswing is already operating with key partners including Amazon Web Services, Anthropic, Apple, Broadcom, and Cisco, as well as CrowdStrike and others. The project could expand as safeguards are tested and shown to be fit for public use.

Via Reuters