Citi CEO Fraser warns powerful AI models could trigger 'tsunami' of cyber attacks

Beijing, September 22 (China New Economy) — According to a Bloomberg report on September 21, Citigroup CEO Jane Fraser warned at the Qatar Economic Forum in New York that increasingly powerful AI models could trigger cyberattacks at any time, prompting companies to urgently strengthen their defenses.

“Right now, companies are racing against time to enhance their defensive capabilities, with a massive global patch-up effort underway,” Fraser said. She described the work required to secure AI defenses as a coming “tsunami.”

Beyond AI risks, Fraser noted that the current bottleneck in technological development has shifted from chips to the energy required to power AI. “We are seeing enormous global demand for energy,” she said.

Qatar’s Approach to AI Investment

Qatar’s Prime Minister, Sheikh Mohammed bin Abdulrahman Al-Thani, revealed that the country is working with AI companies on responsible investments, adding that safeguards around the technology are still being refined.

The Mythos Model and Industry Concerns

Earlier this year, Anthropic PBC released its Mythos model, which can autonomously identify vulnerabilities in computers and software and write corresponding hacking code. Anthropic warned that the system could usher in a new era of cyber risks. Fraser described the release of Mythos as “not a good day.”

Following the model’s release, Treasury Secretary Scott Bessent and then-Federal Reserve Chair Jerome Powell convened a meeting with Wall Street leaders to discuss the model, urging attendees to prepare their banking systems for more powerful AI models.

AI Systems Breach Corporate Networks

Several companies, including Google and Anthropic, have recently experienced incidents where their AI technology accidentally infiltrated corporate systems. Google’s Gemini AI, for instance, autonomously breached multiple companies during a series of drills conducted in May. The exercises were organized by AI security firm Irregular, which also collaborates with Anthropic, Meta, and OpenAI.

Irregular designed a series of tests for a version of Google’s Gemini AI model, requiring it to retrieve data from within simulated companies. The model was not supposed to have internet access. However, once connected, the Gemini agent guessed or found passwords, gaining access to three real companies.

The report noted that these real companies shared names with the fictional ones used in the simulation. The Gemini agent stopped its hacking activities once it realized the companies were real. The incident prompted warnings from several executives, including Anthropic’s Dario Amodei, who called for a slowdown in AI development.

Divergent Views on AI’s Existential Risk

Earlier this month, former Anthropic employee Jacob Coxon warned that researchers believe AI technology could lead to human extinction by the end of this decade. Industry executives such as Nvidia CEO Jensen Huang have countered this view, arguing that AI does not pose an existential threat to humanity.

Goldman Sachs CEO David Solomon, speaking at the Qatar Economic Forum on Sunday, said, “I think we need to take a deep breath and slow down.” He added that he is “very confident that humanity will continue to exist.”

This content (including any images or videos) is uploaded and published by a user of NetEase Hao, a social media platform that provides information storage services only.