Goldman Sachs’s chief executive, David Solomon, has said he is “hyper-aware” of the capabilities of Anthropic’s Mythos AI model and is working “closely” with the tech firm after it issued warnings about the cybersecurity risk it poses.
The US bank had been monitoring the rapid advances in artificial intelligence, including large language models (LLMs), as part of wider efforts to protect itself from hackers.
“Obviously the LLMs are making rapid progress and we’re hyper-aware of the enhanced capabilities of these new models with the help of the US government and the model publishers,” Solomon told analysts on an earnings call on Monday.
That included Anthropic, the company behind the Claude family of AI tools. Last week it claimed that its latest model, Mythos, posed an unprecedented risk because of its ability to expose flaws in IT systems.
“AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities,” Anthropic said in a blogpost last Wednesday. “The fallout – for economies, public safety, and national security – could be severe.”
Solomon said on Monday: “We’re aware of Mythos and its capabilities … We have the model. We’re working closely with Anthropic and all of our security vendors to kind of harness frontier capabilities wherever it’s possible. And this will continue to be an important focus.
“We are very focused on supplementing our cyber and infrastructure resilience. And this is part of our ongoing capabilities that we have been investing in, and are accelerating our investment in.”
The news comes after the US Treasury secretary, Scott Bessant, summoned Solomon and other big American bankers to Washington to discuss the Mythos model last week.
That meeting focused on heads of so-called systemically important banks – where regulators believe that a major disruption to their operations, or their potential collapse, would put financial stability at risk.
On Monday the UK government’s AI Security Institute (AISI) warned that Mythos was a “step up” over previous models in terms of the cyber threat it posed.
AISI said Mythos could carry out attacks that required multiple actions and discover weaknesses in IT systems without human intervention. It said these tasks would normally take human professionals days to carry out. Mythos was the first AI model to successfully complete a 32-step simulation of a cyber-attack created by AISI, solving the challenge in three out of its 10 attempts.
AISI said Mythos appears to be capable of autonomously attacking small, weakly defended IT systems but it could not say for sure whether it could attack well-defended systems because its tests lack security features, such as defensive tools.
The AISI blogpost ended with a warning that future advanced AI models will only improve on Mythos, so “investment now in cyber defense is vital”.
UK regulators are due to raise the issue of Mythos’s risks with British bank bosses and government officials in the coming weeks. The Cross Market Operational Resilience Group (CMorg), made up of chief executives as well as officials from the Treasury, Bank of England, Financial Conduct Authority and National Cyber Security Centre, are due to meet within the next fortnight.
The Bank of England, which is handling communications regarding Cmorg, declined to comment.
