Lagarde Called for Cold War-Style AI Rules for Finance

The world's most powerful financial regulators just stopped arguing about whether AI will take your job and started arguing about whether it can blow up your bank. In Venice on June 17, European Central Bank President Christine Lagarde warned that artificial intelligence could trigger financial crises worse than any technology-driven job loss — and called for global AI governance modeled on Cold War nuclear non-proliferation treaties. Days earlier in Brussels, IMF Managing Director Kristalina Georgieva said frontier models like Anthropic's Mythos "can be used to destroy the financial system." The ECB has already stress-tested 109 eurozone banks against a severe AI-powered cyberattack. Lagarde is now writing directly to bank CEOs. The weapons exist. The treaties do not.

What's new: Lagarde's Venice speech is the first time a sitting ECB president has compared AI governance to nuclear arms control — and framed financial instability, not unemployment, as the primary systemic threat.

• "In recent history, there is one force that has destroyed more jobs and wiped out more savings than technology ever has, and that force is financial crises," Lagarde said.
• She warned AI will "reshape the financial sector from within, creating new concentrations of risk and new openings for those who would do harm."
• Georgieva, presenting the IMF's eurozone assessment in Brussels, said Mythos-class models can patch vulnerabilities "with speed and scale that was unthinkable until now" — but "in the wrong hands, that same capacity can be used to destroy the financial system."

Why it matters now: Washington just demonstrated it can switch off frontier models overnight. Europe discovered it cannot access the defensive tools those models provide. Markets are pricing trillion-dollar AI IPOs while regulators admit they have no global framework to govern dual-use capability.

• Georgieva: "What we recognize is that Mythos is just the beginning, there will be more like it."
• She flagged a separate risk — an AI investment bubble bursting with "low probability, very high impact" consequences for financial stability.
• Federal Reserve Vice Chair for Supervision Michelle Bowman told an FSOC roundtable in May that Mythos "accelerates the process of detecting cyber vulnerabilities" — useful for defense, catastrophic if weaponized.

What the ECB did: Lagarde is not waiting for treaties.

• The central bank simulated an extreme AI-driven cyberattack across 109 banks.
• Most weaknesses exposed in the exercise have been remedied, per Lagarde.
• She will contact bank CEOs directly to demand that cyber resilience be treated as major capital expenditure — not compliance theater.
• "We cannot stop artificial intelligence, even with our sound regulations," she said. "What we can do, however, is prepare ourselves."

The governance gap: Lagarde's non-proliferation analogy is deliberate. Nuclear treaties worked because adversaries accepted shared limits through a formal architecture. AI has no equivalent.

• Georgieva conceded: "There is no world cyber security organization and in the current geopolitical environment it's hard to imagine that we can have one."
• The Financial Stability Board has called for stronger safeguards. Euro-area finance ministers have sought access to Mythos for defensive testing. Washington suspended that access globally on June 12.
• Lagarde urged a European capital markets union alongside tighter oversight — arguing the continent must build financial resilience while negotiating access to tools controlled abroad.

The market read: This is not abstract central-bank rhetoric landing in a vacuum.

• Oracle lost a quarter of its market value after pledging $70 billion in AI capex — proof investors are starting to price infrastructure bills, not just revenue dreams.
• OpenAI and Anthropic have both filed confidential IPO paperwork. Georgieva's bubble warning lands as public markets prepare to absorb trillions in new AI equity.
• Advanced economies, Georgieva said, "have to figure out a way to help the developing world" defend against frontier-model risks — because "if there is a big weakness, it will be utilized, and the world is integrated, the financial systems are integrated."

Convina's view: Lagarde and Georgieva are telling the truth the AI industry keeps dodging: the same models racing toward IPO are dual-use financial infrastructure, and nobody has built the governance layer to match. Job-loss anxiety makes headlines; balance-sheet contagion moves central banks. Europe's Mythos access fight and Washington's export-control kill switch prove the asymmetry — offensive capability proliferates through markets, defensive access gets rationed through geopolitics. Boards pricing AI vendor risk should add a line item regulators are now explicit about: not just data leakage or model drift, but the probability that the tools securing — or attacking — your payment rails sit behind a sovereign off-switch no procurement contract can override. Lagarde wants nuclear treaties. What she has is a letter to bank CEOs and a stress test that already found holes. That is not preparation. That is a countdown.