Goldman Sachs CEO 'Hyper-Aware' of Cybersecurity Risks from Anthropic's Mythos AI Model
Key Takeaways
- ▸Anthropic's Mythos AI model has demonstrated unprecedented cybersecurity risks by autonomously discovering and exploiting software vulnerabilities at a level exceeding most human experts
- ▸Goldman Sachs and other systemically important banks are actively working with Anthropic to understand and defend against Mythos's capabilities while accelerating cyber resilience investments
- ▸UK regulators and US Treasury officials have elevated Mythos to a high-level policy concern, summoning bank leaders and scheduling briefings on the model's potential impact on financial stability
Summary
Goldman Sachs Chief Executive David Solomon has stated he is "hyper-aware" of the cybersecurity capabilities and risks posed by Anthropic's latest AI model, Mythos, which has raised alarm among financial regulators and government officials worldwide. Solomon revealed that Goldman Sachs has access to the Mythos model and is working closely with Anthropic and security vendors to understand and mitigate its capabilities. Anthropic warned last week that Mythos represents an unprecedented cybersecurity risk due to its ability to discover and exploit software vulnerabilities at a level that surpasses most human experts.
The UK's AI Security Institute (AISI) corroborated these concerns, describing Mythos as a "step up" from previous models in terms of cyber threats. Testing showed that Mythos could autonomously execute complex, multi-step cyberattacks and identify IT system weaknesses without human intervention—tasks that typically require days for human professionals. AISI confirmed that Mythos was the first AI model to successfully complete a 32-step simulated cyberattack scenario, solving it in three out of ten attempts. The revelations prompted US Treasury Secretary Scott Bessent to summon major American bank chiefs to Washington last week, and UK regulators are preparing to brief British bank leaders on the risks in coming weeks.
- Mythos successfully completed a 32-step simulated cyberattack—a first for AI models—raising concerns that future advanced models will pose even greater threats to financial infrastructure
Editorial Opinion
While Anthropic's transparency about Mythos's capabilities is commendable and necessary for informed policymaking, the incident underscores a critical tension in AI development: the race to deploy advanced frontier models may be outpacing the financial sector's ability to defend against them. Goldman Sachs's pragmatic approach—gaining access to the model and collaborating with Anthropic on defenses—represents a reasonable interim strategy, but it highlights the urgent need for industry-wide standards and government-led cyber resilience frameworks tailored to advanced AI threats.
