• 2 min read
Jamie Dimon warns Anthropic Mythos is too risky to open
JPMorgan CEO Jamie Dimon says Anthropic’s Claude Mythos poses serious security risks, comparing broad access to handing out ballistic missiles.
Image: TechRadar
JPMorgan CEO Jamie Dimon has warned that broad access to Anthropic’s Claude Mythos could pose major security risks, calling it a “real issue” and comparing release of the model to “giving ballistic missiles to individuals.” He made the remarks at the Pennsylvania Defense and Innovation Summit, where he pointed to the system’s ability to find cybersecurity flaws and generate working exploits.
The concerns come after the US government instructed Anthropic to block access to foreign nationals, citing security worries. According to the source, Claude Mythos is currently limited to a small group of organizations, companies, and federal and military departments, while public access remains blocked.
TechRadar notes that Mythos is not a science-fiction threat like Skynet, but it has already demonstrated capabilities that could be dangerous if misused. The model can:
- detect cybersecurity vulnerabilities
- report on those vulnerabilities
- generate automated exploits
That makes it useful for defensive security work, but it could also help attackers scale ransomware, phishing, and other data-driven scams.
Recommended reading
Kimi K3 arrives with 2.8T params and pricey tokens
The reported risks go beyond cybersecurity. TechRadar says the model has also shown powerful capabilities in scientific research, particularly biology, raising concerns that attackers could use it to simplify and misuse sensitive knowledge.
Access restrictions and future availability
Despite the current limits, Anthropic has said it wants Mythos-level features and capabilities to become more widely available. Before that happens, the company would need to build and apply stronger safeguards, especially around cybersecurity and biological or medical tasks.
TechRadar suggests the most likely outcome is eventual broader access with restrictions on the highest-risk functions rather than a full public release. It also points to Project Glasswing, which is already operating with a small group of partners including Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, and CrowdStrike. That effort is expected to expand until those safeguards are proven fit for public use.
Reuters was cited by the source.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via TechRadar
