Goldman Sachs' Chairman is 'Hyper-Aware' of the Hazards from Anthropic's Mythos AI

Goldman Sachs' CEO, David Solomon, has stated that he is 'hyper-aware' of the capabilities of Anthropic's Mythos AI model and is working 'closely' with the tech business following warnings about the cybersecurity danger it poses.

The US bank had been tracking significant breakthroughs in artificial intelligence, especially large language models (LLMs), as part of a larger attempt to safeguard itself against hackers.

Solomon indicated that LLMs are progressing quickly, noting the improved capabilities of new models supported by the US government and publishers during an earnings call.

Anthropic Identifies Mythos as a Major Cyber Risk

That includes Anthropic, the company responsible for the Claude series of AI technologies. Last week, it claimed that its latest model, Mythos, posed an unprecedented risk due to its capacity to reveal faults in IT systems.

According to Anthropic, AI models now possess advanced coding abilities, enabling them to outpace all but the most skilled humans in identifying and exploiting software vulnerabilities. The potential consequences for economies, public safety, and national security could be significant.

Besides, Solomon stated that they are aware of Mythos and its capabilities, and are working closely with Anthropic and security vendors to leverage frontier capabilities, maintaining it as a key focus moving forward.

Remarkably, efforts are being made to enhance cyber and infrastructure resilience, with a commitment to increasing investment in these areas. This announcement follows a meeting involving US Treasury Secretary Scott Bessent and major American bankers to discuss the Mythos model.

Regulators Warn of Mythos Threats

That meeting addressed systemically important banks, which regulators identify as critical to financial stability due to their potential operational disruptions or collapse. Additionally, the UK government's AI Security Institute (AISI) indicated that Mythos represents an increased cyber threat compared to earlier models.

Meanwhile, AISI reported that Mythos could autonomously execute multi-step cyber-attack simulations and identify vulnerabilities in IT systems, tasks that typically demand days of effort from human experts. Notably, Mythos was the first AI to complete a 32-step cyber-attack simulation, achieving this in three out of ten attempts.

AISI indicated that Mythos can autonomously target small, poorly defended IT systems, but the effectiveness against well-defended systems remains uncertain due to the absence of defensive features in its tests.

The AISI blog post emphasises the importance of investing in cyber defence now, as future advanced AI models are expected to enhance Mythos.

Further, UK regulators will address the risks associated with Mythos in discussions with British banking executives and government officials soon. The Cross Market Operational Resilience Group (CMorg), consisting of chief executives and representatives from the Treasury, Bank of England, Financial Conduct Authority, and National Cyber Security Centre, is set to convene within the next two weeks. The Bank of England, managing communication for CMorg, has refrained from commenting.

Read more news:

• Britain might Adopt Single Market Regulations without MPs' Approval as Part of the UK-EU Reset
• Google Finance, a New AI-Powered Service, is Spreading to Over 100 Countries
• IEA Chief: Present Energy Crisis is Worse than 1973, 1979, & 2022 Together