Goldman Sachs Chief Says It’s ‘Hyper-Aware’ of Anthropic Mythos Cyber Risks

Goldman Sachs CEO David Solomon says the bank is paying close attention to Anthropic’s new Mythos model, describing Goldman as “hyper-aware” of how quickly frontier AI systems are improving at cyber-related tasks. Speaking on the bank’s April 13 earnings call, Solomon said Goldman is working closely with Anthropic and security vendors as it weighs both the risks and the defensive potential of the model.

That matters because Anthropic itself has framed Mythos as a major step up in offensive and defensive cyber capability. In its April 2026 write-up, the company said advanced AI models can now outperform nearly all human experts at finding and exploiting software vulnerabilities, and it limited access to Mythos rather than broadly releasing it.

Weapons of Math Destruction by Cathy O’Neil

A neutral, readable look at how algorithmic systems can create real-world harm at scale. It fits this story’s broader questions around AI risk, oversight, and unintended consequences.

Check Price on Amazon

Goldman is not reacting in isolation. Reuters reported that Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell met with CEOs from several major U.S. banks in Washington on April 9 to discuss Mythos-linked cyber risks. That meeting underscored how quickly this has moved from a lab safety discussion into a financial-stability and critical-infrastructure issue.

The UK is sounding a similar alarm. The AI Security Institute said Mythos represents a step up over earlier frontier models and noted that it was the first system to complete the institute’s 32-step simulated cyberattack chain from start to finish, succeeding in 3 of 10 attempts. AISI also said the model appeared capable of autonomously attacking small, weakly defended systems, while stopping short of saying it could reliably penetrate well-defended environments.

That distinction is important. The immediate fear is not that every bank network suddenly becomes trivial to crack, but that the cost, speed, and skill threshold for meaningful cyber offense keeps dropping as models get better. For big financial firms, that means AI is no longer just a productivity tool or a software assistant. It is also a new threat multiplier that can pressure incident response, vulnerability management, and resilience planning all at once.

Army of None by Paul Scharre

This book offers useful background on autonomy, military technology, and human oversight. It complements the article by putting frontier AI risk into a broader security context.

Check Price on Amazon

For Anthropic, Mythos is becoming a test of whether an AI company can warn publicly about frontier risk while still working with major enterprises that want access to the same tools. For banks like Goldman Sachs, the question is more practical: how do you use leading-edge AI to strengthen cyber defenses without widening the attack surface at the same time? That tension is likely to define the next phase of AI adoption in finance.

For related coverage, see Claude Code lands in Slack for dev workflows, Altman: OpenAI’s Pentagon Deal Was Rushed and “Looks Bad”, and Claude Hits #1 as Pentagon Clash Fuels App Store Surge.