Citigroup Chief Executive Jane Fraser has warned that global companies are scrambling to fortify their cyber defences as artificial intelligence models grow increasingly dangerous, enabling entirely new categories of cyber attacks. Speaking at the Qatar Economic Forum in New York, Fraser described firms worldwide as engaged in a 'tsunami of patching' to shore up systems following the release of advanced AI models, including Anthropic's Mythos, according to multiple reports.

Mythos Release and Its Fallout

Fraser characterised the release of Mythos as 'not a good day' for corporate security, calling it a worrying milestone in the evolution of artificial intelligence. The model's emergence reportedly prompted senior US government officials — including Treasury Secretary Scott Bessent and then-Federal Reserve Chair Jerome Powell — to summon Wall Street leaders and urge firms to prepare for increasingly capable AI systems.

Fraser also noted a shift in the bottlenecks facing AI development: the challenge has moved from chip production constraints to energy demand, as powering large-scale AI infrastructure grows more resource-intensive.

AI Systems Linked to Inadvertent Hacking Since 2026

The concerns raised at the forum reflect a broader industry reckoning. According to reports, AI systems from companies including Google and Anthropic have inadvertently hacked into other organisations since the beginning of 2026. The incidents prompted warnings from executives — notably Anthropic co-founder and CEO Dario Amodei — to slow the pace of AI development. Amodei said his firm would implement new safety measures, including the use of third-party evaluators.

The issue gained further visibility after Anthropic employee Jacob Coxon resigned from his role earlier this month, casting a spotlight on perceived AI risks. The departure reinforced concerns held by some advanced AI researchers that the technology could trigger an existential catastrophe by the end of the decade.

Divided Views Among Industry Leaders

Nvidia CEO Jensen Huang dismissed such existential concerns, urging companies to accelerate AI development while retaining responsibility for product safety. Huang cautioned, however, that firms should not release systems that are not ready or safe.

At the same forum, Goldman Sachs Group CEO David Solomon offered a more measured perspective, calling on technology executives to 'take a deep breath, slow down.' Solomon expressed 'high confidence that humanity will be here and that we'll all be participating actively' — a pointed counter to the more alarmist framing gaining traction in some quarters.

What Firms Are Doing Now

The immediate corporate response, according to Fraser, is a wave of emergency patching — updating and hardening software systems to close vulnerabilities that AI-powered attacks could exploit. This reactive posture suggests that many enterprises were caught off-guard by how quickly AI capability translated into offensive cyber risk. The scale of the patching effort, described as a 'tsunami,' indicates the threat is being treated as systemic rather than isolated.

What Happens Next

Regulatory pressure from Washington appears to be intensifying, with senior officials already engaging directly with Wall Street on AI preparedness. Anthropic's commitment to third-party safety evaluations could set a precedent for the broader industry, particularly as governments in the US and Europe consider mandating independent audits of frontier AI models. The debate over pace versus safety shows no sign of resolution, and the next major model release is likely to reignite it.