Finance Ministers and Bankers Anxious Over Anthropic's AI Model Claude Mythos
Several finance ministers, bankers, and global financial industry players are reportedly growing anxious about Anthropic’s latest artificial intelligence (AI) model, named “Claude Mythos”.
This model is considered to possess capabilities that could potentially disrupt financial system security.
These concerns have emerged in various international forums. Even the BBC reported that Mythos was the main topic of discussion at the International Monetary Fund (IMF) meeting in Washington DC, held from 13-18 April.
Canadian Finance Minister François-Philippe Champagne described the issue as serious enough to garner attention from finance ministers worldwide.
“The difference is, (with the) Strait of Hormuz (conflict), we know its location and scale. Meanwhile, the issue with Anthropic is something that is not yet fully understood,” he added, attempting to illustrate the urgency of the matter.
He emphasised the need for safeguards and procedures to ensure the resilience of financial systems remains intact.
According to him, the main challenge posed by this technology is its nature, which is not yet fully comprehended.
Unlike conventional risks, the capabilities of AI like Mythos are seen as presenting threats that are not yet clearly mapped out.
The Claude Mythos model itself is part of Anthropic’s Claude AI system, developed as a competitor to ChatGPT and Gemini.
This model is described as the most capable one ever created by Anthropic, with standout abilities in cybersecurity.
Mythos is said to be able to discover old bugs or easily exploit system vulnerabilities. These capabilities are the primary source of concern.
Mythos is reportedly capable of identifying security gaps in various systems, even potentially simulating and executing large-scale cyber attacks automatically.
These concerns have also prompted responses from the financial industry. Barclays CEO CS Venkatakrishnan stated that the risks are serious enough to require immediate deeper understanding, including any vulnerabilities that may be exposed.