Anthropic’s Mythos: The AI model ‘too dangerous’ to release?

At a high-profile event featuring JP Morgan Chase CEO Jamie Dimon and Anthropic CEO Dario Amodei, a stark warning was issued about the double-edged sword of the company’s latest AI model, Mythos. 

The new model has unearthed tens of thousands of software vulnerabilities, including decades-old bugs in critical infrastructure. To illustrate the jump in power, an earlier model found 20 bugs in the Firefox browser, whereas Mythos found nearly 300.

Amodei warned that since geopolitical adversaries like China are roughly 6 to 12 months behind this technology, the West has a very narrow window to patch these flaws before they are exploited. 

Amodei emphasizes the risk of an enormous increase in breaches and ransomware attacks targeting hospitals, schools, and global banks. Because the model is so potentially dangerous in the hands of criminals, Anthropic has limited Mythos to only a few select partner companies.

“The danger is just some enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that’s done from ransomware on schools, hospitals, not to mention banks,” Amodei said.

Alongside Jamie Dimon, Anthropic unveiled 10 new AI agents designed to automate complex investment banking and back-office tasks. 

The event signaled Anthropic’s growing dominance over OpenAI in the corporate sector, specifically within the financial industry as both company's eye potential IPOs. Anthropic also announced that its latest public model, Claude Opus 4.7, now leads benchmarks for financial analysis.

Anthropic has announced the development of Claude Mythos, a frontier model with such advanced cybersecurity capabilities that the company has deemed it a public safety risk. 

 For that purpose, previously key representatives from British banks, insurers, and stock exchanges are set to brief on these risks within the next two weeks. Anthropic describes the model’s deployment as part of a controlled initiative where select organizations use the AI for defensive cybersecurity purposes.

Amodei compared AI regulation to the car industry, stating, “You can’t just start a car company without “Are there brakes on this thing.”

While the current moment is dangerous, Jamie Dimon described the risk as a transitory period, and Amodei noted that once the bugs are found and fixed, the world will be more secure.