Regulators in Five Jurisdictions Are Watching Anthropic's Mythos Over Bank Risks

The list of watching regulators keeps growing: Australia's Securities and Investments Commission (ASIC) has officially confirmed that it is monitoring the development of Mythos, a new language model from Anthropic. The move comes amid a growing international reaction: within a few weeks, isolated concern has turned into coordinated global scrutiny.

Before ASIC, the Bank of England, the US Federal Reserve and the US Treasury had already said they were monitoring Mythos. ECB President Christine Lagarde added a key warning to the chorus of concern: there is as yet no governance framework for the new model. For the financial architecture of a eurozone with a GDP of more than 14 trillion euros, those are not just words.

What is Mythos, and why did it provoke such a reaction? Anthropic, best known for its Claude family of models, is developing Mythos as a system with expanded capabilities for autonomous action in the financial environment. According to available information, the model can interact with banking infrastructure, process transactions and make financial decisions with minimal human involvement. That is precisely what drew the attention of supervisors: autonomous AI in the financial sector carries a fundamentally different level of risk than familiar analytical tools.

Regulators are worried about several scenarios. The first is systemic risk: if a powerful AI model becomes widespread in banking and makes a mistake, the consequences could spread across the whole market, because many institutions would be acting on the same algorithm. The second is a lack of transparency: how large language models work is hard to verify, which means a regulator cannot guarantee compliance with established rules. The third is the pace of change: AI companies release new versions faster than standards can form.

Anthropic positions itself as a safety-first company. The philosophy of constitutional AI and public commitments to responsible development are part of its corporate identity. Still, the regulators' reaction shows that safety declarations and a system's real readiness to operate in critical infrastructure are different things. Supervisors want to see concrete mechanisms for auditing, control and emergency shutdown.

The episode fits a broader trend. After the boom of 2023–2024, when AI tools began to penetrate financial services, regulators around the world are moving from observation to active measures. In Europe, the AI Act sets requirements for high-risk systems. In the US, Congress is working on bills, while the SEC, CFTC and OCC are shaping their own approaches. Australia traditionally follows international standards of financial regulation and, judging by ASIC's statement, does not intend to fall behind.

For companies building AI for the financial sector, a new reality is setting in: a model's capabilities no longer define the horizon of its use. Regulatory acceptability is now a constraint just like technical readiness. Mythos may be an arbitrarily powerful tool, but without clear frameworks of accountability and alignment with supervisors, its commercial deployment in banking remains in question.

For Anthropic, this is a moment of testing. The company built its reputation on the principle of safety first, and now it must prove that those principles work not only in the lab but also under the close scrutiny of the five largest financial regulators in the world.