'Only 12 Months to Act' and Resolve Vulnerabilities: Anthropic's Warning Concerns Governments and Banks

Artificial intelligence may have opened a critical phase for global cybersecurity. During a company event, Dario Amodei, CEO of Anthropic, stated that the new AI model named Mythos has identified tens of thousands of software vulnerabilities, some of which date back decades and are present in systems considered fundamental.

According to Amodei, this discovery creates a limited time window for tech companies, governments, and financial institutions, which are called to act quickly to mitigate risks. The geopolitical context plays a relevant role: models developed in China are reportedly six to twelve months behind Mythos, offering a time advantage to address the issue.

Anthropic highlights how the ability of its models to identify vulnerabilities has rapidly increased. A previous system was able to identify about 20 flaws in the Firefox browser, while Mythos has detected nearly 300 in the same area. On a larger scale, the total number of vulnerabilities detected now stands in the tens of thousands.

Much of this information has not been made public. The choice stems from the fear that disclosure could facilitate the activities of malicious actors before vulnerabilities are corrected. Amodei emphasized that the main risk lies in a significant increase in ransomware attacks and breaches, with potential impacts on critical infrastructure such as schools, hospitals, and financial institutions.

Despite the outlined scenario, the company leaders expressed a cautious optimism. Amodei described the current phase as a 'danger moment' but also an opportunity to structurally improve software security, emphasizing that while the number of vulnerabilities is high, it remains finite and thus manageable.

This stance is shared by Jamie Dimon, CEO of JPMorgan Chase, who was present at the event. Dimon characterized the risks related to AI as part of a transitional period, destined to evolve with the adoption of new countermeasures and technologies.

During the event, Anthropic announced the expansion of its platform dedicated to financial services. The offering includes ten new AI agents designed for investment banking activities and back-office operations, along with integration with tools from the Microsoft Office suite.

At the same time, the company highlighted the performance of the Claude Opus 4.7 model, currently available, which would rank at the top of benchmarks for financial analysis.

Touching on the regulatory front, Amodei concluded by proposing an approach to AI regulation similar to that adopted in the automotive industry: a balance between safety and innovation, with targeted controls on the most relevant criticalities without slowing technological development.