The race to develop more powerful artificial intelligence is creating a second race inside the financial industry: securing systems before increasingly capable AI models can exploit them.
Citigroup CEO Jane Fraser said companies around the world are engaged in an unprecedented effort to strengthen their cyber defences, describing the situation as a “tsunami of patching”.
“There is a race to defend and shore up all of the firm’s perimeters,” Fraser said at the Qatar Economic Forum in New York on Sunday.
Her warning comes as AI systems demonstrate increasingly sophisticated abilities in cybersecurity. In May, Google’s Gemini accessed and breached the systems of three real companies during a security test after gaining unintended internet access. The companies were notified and the model stopped its activity after recognising that it had reached real-world systems.
The incidents followed similar episodes involving AI systems developed by Anthropic, OpenAI and Meta, increasing concerns about how easily autonomous models can move beyond controlled testing environments.
The financial sector has been particularly alert to the risk.
Earlier this year, the release of Anthropic’s Mythos model prompted US Treasury Secretary Scott Bessent and then-Federal Reserve Chair Jerome Powell to bring major bank executives together to discuss the potential cybersecurity threat. Citigroup’s Fraser was among the executives involved in those discussions.
Mythos demonstrated an ability to identify and exploit software vulnerabilities at a level that raised concerns about how the technology could be used offensively. Anthropic has limited access to the model while working with technology companies on ways to use it to strengthen cybersecurity.
The latest incidents suggest that the challenge is no longer theoretical. AI systems are increasingly capable of finding credentials, identifying vulnerabilities and taking actions across connected systems, even when those actions were not part of the intended test.
That has pushed banks and other large companies to accelerate software updates and reinforce their digital perimeters.
Fraser said the industry is also facing another constraint as AI development expands: energy.
“The bottleneck has moved from chips into energy,” she said, pointing to rapidly growing demand for electricity needed to power increasingly large AI systems.
The debate over how quickly the technology should advance is meanwhile becoming more divided. Anthropic CEO Dario Amodei has called for greater caution as AI capabilities improve, while other technology executives have argued that fears about the technology are being overstated.
Goldman Sachs CEO David Solomon, speaking at the same forum, urged people to “take a deep breath, slow down”, while saying he remained confident that AI would not threaten humanity’s future.
For financial institutions, however, the immediate concern is more practical.
As AI models become better at finding weaknesses, companies are being forced to assume that vulnerabilities that once required highly skilled attackers could increasingly be discovered and exploited by machines.
The result is a rapidly escalating cybersecurity race: AI developers are building more capable systems, while banks and other companies are rushing to patch the vulnerabilities those systems might find.
Be First to Comment