Swiss bankers wary, but not spooked by Mythos AI threat

The jury is still out on whether Mythos will harm or help banks. Keystone

The Swiss financial sector is aware of the potential threat posed by the latest artificial intelligence (AI) frontier models, such as Anthropic’s Mythos. But the authorities are unwilling to announce red alert until more is known about the technology.

Mythos is deemed so powerful that it has been held back from general release by United States AI company Anthropic. The model is reportedly capable of swiftly spotting previously undiscovered IT bugs.  

This presents a double-edged sword. The technology could help strengthen cyber defences by highlighting weaknesses. But it also raises fears that it could be used to unleash a wave of cyberattacks on critical infrastructure, such as financial systems.

+ Cyberwar attacks heart of European infrastructure 

“From today’s perspective, the uncontrolled and immediate availability of AI models like Mythos would be classified as a systemic risk,” the Swiss Financial Market Supervisory Authority (FINMA) told Swissinfo.ch. 

“In such a scenario, virtually all existing software systems could be simultaneously affected by a multitude of previously unknown zero-day vulnerabilities, which would be exploited immediately using AI.” 

‘Project Glasswing’ available to few

“However, current indications suggest that these capabilities will be made available in a controlled and phased manner.” 

 Anthropic has limited early access to Mythos to a handful of US companies, such as Amazon Web Services, Microsoft, Nvidia, the Linux Foundation and JP Morgan. The select group of multinationals in the ‘Project Glasswing’ initiative are testing Mythos on their own systems in a controlled environment. 

The US and Britain have reportedly convened emergency meetings of bankers and politicians to address their concerns about the risks posed by the technology. Financial executives and regulators from many countries have expressed concerns about the reported power of Mythos.  

In Switzerland, “There have so far been no special political or crisis meetings dedicated to Mythos or similar large language models,” the Swiss Bankers Association (SBA) told Swissinfo.ch. “Instead, the topic is being addressed through existing structures that deal with cyber security and operational resilience.” 

The Swiss National Cybersecurity Centre (NCSC) is coordinating Switzerland’s response to Mythos and is currently making efforts to find out more about the AI model and Project Glasswing.   

‘Dramatic portrayals’ amid real threats 

The agency has adopted a cautious approach amid suspicions that Anthropic is exaggerating the capabilities of its new product ahead of a potential stock market listing. “The sometimes dramatic portrayals are also influenced by marketing,” the NCSC told Swissinfo in a written statement.  

A full assessment of Mythos is impossible with so few technical details available about the tightly guarded AI model. The NCSC says that AI in general is “less of a game-changer in terms of new risks and more of an accelerator of existing threats”. 

Join the debate:

“Carrying out complex attacks will require fewer resources in future and will thus be accessible to a larger number of actors,” it stated. “As a result, highly complex attacks that were previously only possible for attackers with substantial resources may occur more frequently.” 

The SBA is advising banks to double down on strengthening IT and risk management systems while employing robust oversight of AI models used in operations. “From our perspective, the threat is real, and a pure ‘wait-and-see’ approach would not be sufficient,” the SBA said. 

“Adversaries can use AI to identify vulnerabilities and to develop exploitation techniques more efficiently. This expected increase in speed and complexity reinforces the need to continuously improve processes, maintain a high level of cyber awareness, and adjust tools and capabilities accordingly.” 

Europe fears US dominance on AI 

European countries fear being left behind as the US and China harness the power of AI more effectively.  

Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations, a US foreign policy think-tank, applauds US efforts to tighten cyber defences and maintain a lead in AI.  

“These efforts are urgent because China will develop a model as capable as Mythos soon. China’s best AI models currently lag behind leading US models by about seven months, potentially slightly more,” he wrote in the Financial Times. 

Switzerland and other European countries are trying to close the gap by getting a seat at the table of Project Glasswing. 

“All relevant institutions should have access to this technology to avoid distortions of competition,” German central bank president Joachim Nagel told the International Economic Symposium in Rome last week. 

His views are echoed by Florian Tramèr, a professor of cyber security at the Swiss federal technology institute ETH Zurich.

“Anthropic appears to be pursuing a highly US‑centric security logic, with Europe playing only a minor role,” he wrote in an article published on the ETHZ website. “Models of this kind are potentially relevant to national security, intelligence services and the military. If access remains limited to US stakeholders, these stakeholders will have a head start for some time.” 

Switzerland’s largest bank, UBS, declined to comment on Mythos.

Edited by Veronica De Vore