Anthropic Mythos AI: Jamie Dimon warns broad access is like 'ballistic missiles'

JPMorgan Chase CEO Jamie Dimon has warned that broad access to Anthropic's Mythos AI model poses serious risks, comparing it to 'giving ballistic missiles to individuals.' The remarks, made at the Pennsylvania Defense and Innovation Summit, highlight the tension between advanced AI capabilities and national security. For AI builders, this signals that powerful models with cybersecurity and biology research implications will face tight access controls and rigorous safeguards before any public release.

What happened

Speaking at the Pennsylvania Defense and Innovation Summit, Dimon described the risks posed by Anthropic's Mythos AI as a 'real issue' and said that without proper controls, 'you're giving ballistic missiles to individuals with Mythos, basically.' The US government has already limited access to the model, which is currently restricted to a small set of organizations, companies, and federal and military departments. Public access has been blocked worldwide due to security concerns.

Anthropic has stated it intends to make Mythos-level features more widely available, but only after developing and applying various safeguards. Project Glasswing, a collaboration with partners including Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, and CrowdStrike, is already operating to test these safeguards. The project will likely be expanded until the safeguards are demonstrated to be fit for public use.

Why AI builders should care

For teams building AI products or deploying models in sensitive domains, the Mythos situation is a case study in how advanced capabilities trigger governance requirements. The model's ability to identify and exploit cybersecurity vulnerabilities and generate automatic exploits makes it a powerful defensive tool for white hat analysts, but equally dangerous in the wrong hands. Similarly, its capacity to turn complex biological concepts into actionable information raises dual-use concerns.

Any future public access will likely come with strict feature restrictions, particularly around cybersecurity and biomedical tasks. Builders should expect that accessing Mythos-level capabilities will require partnership programs like Project Glasswing, validated safeguards, and ongoing compliance with government oversight.

Practical implications

AI builders and product teams should start planning for a regulatory environment where powerful models are gated behind verified safety measures. If you are building applications that could benefit from Mythos-like capabilities, you may need to engage with Anthropic's partner ecosystem or similar trusted access programs. The involvement of major tech companies in Project Glasswing (AWS, Apple, Broadcom, Cisco, CrowdStrike) suggests that collaboration with established infrastructure providers will be a prerequisite for early access.

Additionally, internal risk assessments for cybersecurity and biology-related use cases will become standard before deployment. The path to public access is likely to be gradual, with feature restrictions and continuous monitoring.

Caveats

Dimon's remarks come from a single event and reflect his personal perspective as a banking CEO, not official policy. Specific details about Mythos's capabilities, the exact safeguards under development, and timelines for broader release have not been publicly confirmed beyond what is reported. The research pack does not include independent benchmarks or technical specifications for Mythos. Builders should treat this as a signal of the direction of AI governance rather than a definitive roadmap.