Goldman Sachs Collaborates with Anthropic on AI Security Measures
Goldman Sachs is actively leveraging Anthropic’s advanced artificial intelligence model, Mythos, to enhance the security defenses of the financial sector. The bank’s CEO, David Solomon, recently highlighted these efforts during an earnings call held on a Monday, emphasizing the importance of navigating the rapidly evolving landscape of cyber threats presented by advanced AI technologies.
In his remarks, Solomon expressed concern regarding the potential for Mythos to facilitate more sophisticated cyberattacks, stating, "We’re hyper-aware of the enhanced capabilities of these new models." He elaborated on the ongoing collaboration with Anthropic and various security vendors, indicating a commitment to harnessing these frontier capabilities to bolster the bank’s defenses. This commitment aligns with broader efforts within the industry to address rising cybersecurity threats influenced by AI advancements.
The urgency of this initiative comes in the wake of a recent meeting convened by U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell, which gathered Wall Street leaders to discuss the risks associated with Mythos. Solomon did not consider the meeting extraordinary, framing it as part of a long-standing dialogue regarding cybersecurity risks. "This is something the industry is focused on. It’s something we’re focused on. And there’s nothing new in that focus,” he stated, reflecting a recognition of the ongoing nature of these discussions.
Launched earlier this month, Mythos has limited access to a select group of companies under what Anthropic refers to as Project Glasswing. This initiative aims to utilize Mythos to identify security vulnerabilities before malicious actors can exploit similar functionalities. In its initial weeks of testing, Mythos successfully identified thousands of previously unknown vulnerabilities within major operating systems and web browsers. This capability underscores the model’s significance as a tool for preemptively addressing security flaws.
However, the heightened attention surrounding Mythos stems not only from its ability to detect vulnerabilities but also from its potential misuse. Analysts have observed that Mythos marks a pivotal shift in the cybersecurity landscape, possessing the capacity to autonomously generate functional exploits without requiring human intervention. This automation could facilitate cyberattacks at unprecedented speeds and scales, provoking concern among cybersecurity experts and financial regulators alike.
An independent evaluation conducted by Great Britain’s AI Security Institute assessed Mythos’s capabilities, revealing its potential to execute multi-stage attacks on vulnerable networks and to autonomously discover and exploit flaws. The institute’s tests demonstrated that Mythos could undertake complex intrusion scenarios more efficiently than human professionals, completing a 32-step simulated corporate network attack with some success. However, it did note that the true test of Mythos’s effectiveness against well-defended systems remained inconclusive due to the lack of active defenses in its testing environment.
Goldman Sachs’s investment in AI-driven automation has been ongoing for more than a year. Solomon reassured investors that the bank remains committed to enhancing its cyber resilience, bolstered by collaboration with the U.S. government and model publishers. This focus on cybersecurity is part of Goldman Sachs’s broader initiative to enhance operational efficiency through technological advancements.
In February, Goldman Sachs’s Chief Information Officer, Marco Argenti, shared insights into the partnership with Anthropic, mentioning that embedded engineers had spent six months co-developing autonomous agents for tasks such as trade accounting and client onboarding. This suggests that the bank is looking into automating additional facets of its operations, further emphasizing the pivotal role of AI in modernizing financial processes.
Addressing concerns during the earnings call, Solomon acknowledged the inherent risks associated with accelerating technology adoption. He stated, "Whenever you have acceleration of your technology, there are going to be bumps, and there are going to be risk issues." However, he remains optimistic about the transformative power of AI, highlighting its potential for streamlining processes, increasing efficiency, and ultimately driving growth. "I can’t find a CEO that’s not talking about that," he concluded, underscoring a collective recognition of the imperative to navigate the complexities of technological advancements safely.
As Goldman Sachs continues to forge ahead in its AI initiatives, the collaboration with Anthropic serves as a crucial element in fortifying the bank’s defenses against an evolving cyber landscape. This proactive approach reflects a broader industry trend, as financial institutions recognize the importance of integrating cutting-edge technology while addressing the accompanying challenges.