Jamie Dimon Warns Anthropic’s Mythos AI Has Increased Cybersecurity Risk Tenfold

JPMorgan Chase CEO Jamie Dimon has issued a fresh warning about the cybersecurity implications of increasingly powerful artificial intelligence, saying the arrival of Anthropic’s Mythos model has dramatically increased the level of cyber risk.

Speaking in an interview with Bloomberg TV, Dimon said cybersecurity was already one of JPMorgan’s biggest concerns before Mythos appeared. In his view, the emergence of the advanced AI system has made the situation significantly more serious.

Jamie Dimon Says Cyber Risk Has Risen Sharply

Dimon previously described cybersecurity as one of the biggest risks facing JPMorgan and the wider financial sector. He now believes that risk has increased substantially because advanced AI systems can discover software weaknesses at a speed and scale that traditional security teams cannot easily match.

According to Dimon, AI can uncover vulnerabilities that organizations may not have known existed. That creates a difficult situation: the same technology that can help security researchers find and fix weaknesses can potentially be used by attackers to exploit them.

Dimon emphasized that he does not believe people should simply panic about the technology. Instead, he argued that businesses and governments need to respond quickly and develop practical defenses.

Why Anthropic’s Mythos Is Creating Concern

Anthropic developed Mythos as a highly capable AI system focused heavily on cybersecurity tasks. Rather than being broadly available to the public, access has been restricted because of concerns about potential misuse.

The model has demonstrated an ability to identify software vulnerabilities and assist with cybersecurity research. That makes it valuable to security professionals, but the same capabilities could become dangerous if they were placed in the hands of malicious actors.

Earlier in 2026, Anthropic launched Project Glasswing, giving selected organizations access to advanced AI capabilities so they could identify and address vulnerabilities in important software and infrastructure.

AI Could Change the Cybersecurity Arms Race

One of the biggest concerns surrounding Mythos is the speed at which AI can perform security-related work.

Traditional vulnerability research can require significant time and expertise. Highly capable AI systems can automate portions of the process, potentially allowing defenders to identify weaknesses faster—but also giving attackers the possibility of scaling their operations.

Anthropic CEO Dario Amodei previously warned that advanced AI models had already identified tens of thousands of software vulnerabilities. He argued that defenders need to patch these weaknesses before comparable capabilities become widely available to malicious groups.

This creates a race between finding vulnerabilities and fixing them.

Anthropic Is Expanding Controlled Access

The latest developments show that Anthropic is not simply keeping its strongest cybersecurity models locked away.

The company has announced an expanded Cyber Verification Program that provides vetted cybersecurity professionals with access to advanced models under different levels of safeguards. The program includes defensive security work, authorized penetration testing and specialized testing of critical infrastructure.

Anthropic says its earlier Project Glasswing partners identified at least 129,000 verified software vulnerabilities between April and July, including more than 33,000 classified as critical or high severity. The company also reported thousands of additional vulnerabilities from its own scanning efforts.

Mythos Is Not Being Released as a Normal Consumer AI

Despite the attention surrounding Mythos, it is important to distinguish the system from consumer-facing AI assistants.

Mythos has not been broadly released to the public. Anthropic has instead used controlled programs and vetted organizations to reduce the possibility of misuse.

That approach reflects a growing concern within the technology industry: extremely capable AI models may need different safety measures depending on what they can actually do.

A model capable of writing emails or summarizing documents presents a very different security challenge from one capable of autonomously identifying and exploiting software vulnerabilities.

Governments Are Also Watching the Technology

The potential cybersecurity impact of advanced AI has attracted significant attention from governments and financial regulators.

Earlier this year, U.S. officials warned major financial institutions about the risks associated with advanced AI models. Government restrictions and discussions surrounding access to Anthropic's most powerful systems have highlighted the growing national-security dimension of AI development.

The debate is becoming increasingly complicated because the technology can serve both offensive and defensive purposes.

Banks, technology companies and security researchers can use advanced AI to discover vulnerabilities before criminals do. At the same time, the same capabilities could eventually be used to automate attacks.

What This Means for Businesses

For businesses, the rise of AI-powered cybersecurity tools means that traditional security practices may no longer be enough.

Companies may need to:

• Patch critical vulnerabilities much faster.
• Continuously monitor internet-facing systems.
• Test software using automated security tools.
• Protect sensitive credentials and infrastructure.
• Improve incident-response capabilities.
• Monitor emerging AI-assisted attack techniques.
• Regularly audit third-party software and cloud services.

The speed of vulnerability discovery could become particularly important. A security flaw that once remained unknown for months could potentially be discovered much faster by an advanced AI system.

AI Could Also Strengthen Cybersecurity

Despite the warnings, Mythos is not necessarily bad news for cybersecurity.

The same capabilities that make advanced AI potentially dangerous can also help defenders. Security teams can use AI to analyze code, discover vulnerabilities, prioritize threats and investigate suspicious activity.

Anthropic's controlled-access approach is based largely on this defensive potential. By giving vetted organizations access to powerful models, the company hopes they can use AI to strengthen security before attackers gain comparable capabilities.

Final Thoughts

Jamie Dimon’s latest warning highlights an important shift in the cybersecurity landscape. Advanced AI is no longer just a tool for generating text, writing code or answering questions—it is increasingly capable of performing sophisticated security research.

The biggest challenge may be maintaining the advantage of defenders while preventing these capabilities from becoming easily accessible to cybercriminals.

Anthropic's decision to expand controlled access to Mythos and other advanced models suggests the industry is moving toward a model where powerful AI cybersecurity tools are available, but only under strict verification and safety controls.

For now, Dimon’s message is less about abandoning AI and more about recognizing that its cybersecurity impact is arriving much faster than many organizations expected.