Better trading AI can make the same mistake together
The warning is about correlation, or many systems moving together. In the market simulation, similar frontier models did not act like separate bets. If their shared view was accurate, risk fell. If they shared bad information, the same pattern made risk concentrate.
What the source reports
The authors build a market test with AI model traders of different strength. Their abstract says stronger frontier models act more alike. That can help or hurt. When their shared reasoning is right, adding more models can reduce market risk. When they share the same bad facts, the same similarity becomes dangerous. The plain version is simple: upgrading every trader to the newest similar model may improve each trader while making the whole system less varied. The authors also give a limit. They say it is still an open empirical question whether the same pattern appears in other fields, so keep this to simulated markets.
Original source
- Title
- Why Better Models Can Create Riskier Systems: Evidence from LLM Agents in Financial Markets
- Author
- Ross, So, De Simone, Pozniak & Lo
- Publication
- arXiv
- Date
- Thursday, September 3, 2026
Also covered by the other newsletter.