Goldman Sachs Collaborates with Anthropic on AI Cybersecurity Risks After Model Delay

Goldman Sachs is collaborating with Anthropic to assess cybersecurity risks associated with the AI company's latest model, Mythos. The model, described as highly advanced, was not released publicly last week due to potential vulnerabilities. Instead, Anthropic is providing limited previews to select organizations.

The decision followed concerns that the system's capabilities could allow non-experts to exploit weaknesses in major operating systems. Goldman Sachs, which has access to an early preview, is coordinating with Anthropic and its security vendors to address these threats. Cybersecurity remains a core focus for the bank, with ongoing investments in resilience.

Last week, leaders from major US banks met with Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent to discuss the risks posed by the model. The meeting addressed broader cybersecurity issues in the financial sector. Such discussions between banks and government officials on these topics have occurred for years.

During the bank's first-quarter earnings call, Goldman Sachs stated that it is supplementing its cyber infrastructure with support from the US government and AI model developers. The bank is accelerating investments in these areas. Despite the concerns, the institution expressed optimism about AI's potential to improve processes and efficiency in the medium to long term.

The rollout of advanced AI models like Mythos highlights ongoing challenges in balancing innovation with security in the financial industry. Banks are increasingly integrating AI technologies, which requires enhanced protections against potential exploits. This collaboration aims to mitigate risks while exploring AI's benefits for business operations.

Anthropic's approach of limiting access to the model provides time for thorough testing and safeguards. Financial institutions, including Goldman Sachs, are among the organizations receiving previews to evaluate real-world applications. The incident underscores the need for coordinated efforts between technology developers and regulated sectors.