Goldman Sachs CEO David Solomon has revealed that the bank is in close collaboration with Anthropic following alarming warnings about the cybersecurity threats posed by its latest AI model, Cloud Mythos.
As part of a comprehensive strategy to protect against evolving hacking tactics, the US bank has consistently engaged in in-depth evaluation of large language models (LLMs).
“Obviously LLMs are progressing rapidly and we are very familiar with the increased capabilities of these new models with the help of the U.S. government and model publishers,” Solomon told analysts on an earnings call Monday.
According to Solomon, Cloud Mythos AI has cutting-edge capabilities that are known to identify critical cyber vulnerabilities in various operation systems and web browsers.
“We are working closely with Anthropic and all of our security vendors to utilize edge capabilities wherever possible. And this will continue to be an important focus,” he said.
Solomon further assured, “We are very focused on improving our cyber and infrastructure resiliency. And that’s part of our ongoing capabilities that we’re investing in and accelerating our investments.”
Earlier this week, US Treasury Secretary Scott Besant and Federal Reserve Chairman Jerome Powell issued dire warnings to the financial sector, especially bank CEOs, regarding AI-driven cyber risks.
At the meetings, heads of banks were instructed to protect their financial systems against growing AI-driven cyber threats.
Another warning also came on Monday from the UK government’s AI Security Institute (AISI), which said Mythos was “a step up from previous AI models in terms of cyber threats.”
According to AISI, Mythos is highly capable of performing multiple tasks and discovering flaws in IT systems without human oversight.
Leaving human professionals behind, Mythos made history as the first AI model to conquer a 32-step cyber-attack simulation designed by AISI, achieving a success rate of 30% in 10 attempts.
AISI also reported that during testing the Mythos appeared to be capable of autonomously targeting small and vulnerable IT systems. The institute also issued a dire warning that in the near future AI models will only get better at transcending mythos. Therefore investment in cyber defense has become the need of the hour.
Given the unprecedented cyber capabilities of Cloud Mythos, Anthropic has decided not to publicly release this model. Under its newly launched “Project Glasswing” project, Mythos will be available to Google, OpenAI, Microsoft, and other organizations.
