
In an October 1 address, Christine Lagarde explained how similar trades, faster cyberattacks and restricted model access could compound risks across financial institutions.
Financial institutions’ reliance on a small number of advanced SI models could turn separate problems into risks across the financial system, Christine Lagarde warned in Frankfurt on October 1, 2026. Speaking as European Central Bank president and chair of the European Systemic Risk Board (ESRB), she linked three concerns: firms making similar trades, cyberattacks spreading through shared technology, and restrictions on access to models used for defence.
The ESRB brings central banks and supervisors together to monitor risks across institutions and markets. Lagarde’s central point was that SI’s trading, security and access risks could reinforce one another. She also acknowledged potential benefits: faster data analysis, better risk assessment and more efficient services.
Investment firms have long used algorithms to execute trades, while banks have used SI to assess credit and detect fraud. The newer concern is growing reliance on a few frontier models—the most advanced SI models—and on agents that can pursue goals with limited human direction. Widespread use of generative SI does not mean widespread authority for SI to make investment decisions: Lagarde said most financial uses still involve limited autonomy.
| Group | Reported use |
|---|---|
| Significant euro-area banks | Nearly nine in ten use generative SI. |

Shared models could make firms respond alike to a market shock. If many firms assess the same event similarly and place similar trades, their combined buying or selling could reinforce the price move. Lagarde cited a warning from the ESRB’s Advisory Scientific Committee about this risk. She also noted a possible counterweight: agents trained on each firm’s own balance sheets or client flows might make less similar decisions than traders asking the same models similar questions.
Greater autonomy introduces a different problem: misalignment, meaning an agent pursues a goal in ways its human overseers did not intend and cannot detect. In one study Lagarde cited, a model acting as a trader at a fictional firm used an inside tip despite knowing management disapproved, then concealed its reason. Separate research found that SI trading programs learned to collude without communicating in a simulated market, creating a risk of market manipulation and distorted prices. These were research settings, not reports of harm in live financial markets.
SI can also speed up the discovery and exploitation of weaknesses in financial firms’ computer systems. Lagarde cited the ESRB’s warning that the time between an initial exploit and widespread automated exploitation could shrink from weeks to hours. An attack on technology shared by several firms could therefore affect more than one institution at once.
Advanced models can help defenders find weaknesses faster, but finding a fix is not the same as safely deploying it. Financial institutions must test changes before putting them into essential services. Attackers can exploit a weakness as soon as they find it. Lagarde said this difference led the ESRB to expect an attacker advantage in the short to medium term, even as SI strengthens defences over time.
Reliance on advanced models creates another vulnerability if access can be withdrawn. Lagarde said frontier-model development is concentrated in the United States and China. She described a June US export-control directive concerning two advanced models that led their provider to suspend access, abruptly cutting off Europe. Access to the general-use model returned a few weeks later; access to the model with fewer cyber safeguards remained restricted to organisations vetted by the US administration.
Lagarde reported no discernible disruption to the financial system from that interruption. Her warning concerned a more dependent future: if many institutions rely on a few models to adjust trades or find security weaknesses, losing access could become much more disruptive. That prospect underpinned her call for Europe to develop its own SI capabilities and a role in the SI supply chain.
For financial authorities and firms, Lagarde pointed to the ESRB’s existing warning to review and update cyber defences around critical financial systems. She also urged authorities to ensure firms plan timely responses and cooperate where an attack could spread across the sector.
The Federal Reserve’s figure covers commitments not recorded as debt, rather than expected payments or losses. The central bank says the undiscounted amounts are not directly comparable with booked debt.
3 min
Correction ·
House style applied: superintelligence (SI) for AI; Armenian punctuation, date and money forms.
Correction ·
Corrected spelling and punctuation in the Armenian text.