Karsten Wenzlaff, Advisor
August 26th, 2025
Karsten Wenzlaff is a Crowdfunding and Fintech Expert from Germany. He is Secretary-General of Digit[...]
What Happens When AI Agents Start Trading Together?
November 11, 2026 | NCFA Fintech Market Insight | Artificial Intelligence And Data, Capital Markets Infrastructure And Funding, Risk Compliance And Regtech
Financial firms are putting AI agents into production well before giving them broad financial authority. That creates time to set identity, permissions and transaction controls before autonomous execution becomes common.
On November 11, 2026, Bank of England Financial Policy Committee external member Jonathan Hall used his final FPC speech to examine what happens when AI agents get more authority inside financial firms. A Bank of England and FCA survey of 196 firms found that 80% already deploy AI agents and another 14% plan to within 12 months. Only 2% reported agents with broad autonomy, and none of those were in trading and payments. Finance is adopting agents quickly, but firms are still keeping people and hard limits around the decisions that can change money, positions or customer outcomes.
AI Agents Are In Finance But They Are Not Trading Alone
The 2026 Bank of England and FCA survey separates four levels of authority:
- Decision support lets an agent recommend an action while a person decides and executes it.
- Constrained execution lets the agent act inside preset rules.
- Conditional autonomy gives the agent more freedom but still requires approval for some actions.
- Broad autonomy lets it initiate and execute actions with limited real time human involvement.
Among respondents, 80% reported decision support agents, 46% constrained execution and 22% conditional autonomy. Broad autonomy was just 2%. Trade and execution was much lower at 7%, 5%, 1% and 0%.
Payments came in at 7%, 5%, 2% and 0%. Even after planned deployments are included, no more than 9% expect agents within the next year to execute trades, approve customer outcomes, initiate or block payments, or change financial limits and controls.
Canadian firms are already testing that boundary. Questrade MCP lets selected customers give an AI agent access to approved brokerage and market data and prepare an order. The customer still has to approve the trade before it is submitted, so it's not autonomous trading. It's a live example of a financial company giving an agent useful access without handing over final authority.
Shared AI Could Make Financial Firms Sell At The Same Time
Thousands of agents using similar technology can become a market risk. The survey found that the three largest cloud and compute providers account for 79% of reported provider relationships. The three largest model API and hosting providers account for 54%.
Several banks, funds or trading firms could use different applications while relying on the same models, cloud providers or data. If those agents read the same market conditions and respond by selling assets, reducing exposure or pulling back liquidity at roughly the same time, many individually reasonable decisions could hit the market together.
See: AI Agents Enter Governed Financial Workflows
Hall focuses on three things. How concentrated are the firms providing financial services? How similar is their behaviour in normal markets? How similar does it become when markets are under pressure? On whether AI will increase that concentration or correlation, he says “at this point we don’t know, but there is a risk that it will.”
Currently, AI agents are not currently causing market wide instability at scale. The concern comes from growing adoption, common technology providers and the possibility that greater autonomy could make reactions faster and more alike. The Bank of Canada’s 2026 Financial System Survey found that nearly all respondents use AI, mostly at limited or moderate levels, and continue to rely on human judgment for decisions carrying significant financial, legal or reputational consequences.
Project Logos Will Test Whether AI Agents Herd Under Stress
Project Logos is testing the behaviour regulators cannot yet observe at scale in live markets. The BIS Innovation Hub, Bank of England and Deutsche Bundesbank are building a simulated financial market where traditional rules based portfolio managers operate beside LLM based agents. Investor agents allocate capital while both groups make portfolio decisions over time.
Researchers can then quantify whether AI portfolio managers respond differently from traditional automated strategies, whether small changes in models or prompts alter their decisions, and whether their behaviour becomes more correlated when markets deteriorate. Project Logos remains ongoing.
Hall also mentions laboratory research where agents developed unexpected group behaviour, including collusion, and a separate experiment where agents communicated after encountering blocked tasks. Those results show that multi agent behaviour can produce outcomes that were not explicitly planned. They don't however show AI agents colluding in live financial markets.
See: FCA Emerging Technology Horizon Scan 2026
Further, financial market rules already prohibit manipulation, which can make detection harder if agents learn to coordinate indirectly through price, volume, timing or rapid trading. Hall specifically raises tacit collusion and momentum ignition as behaviours that may be difficult to control through permissions alone.
Give AI Agents Only The Financial Authority They Need
Companies don't need to wait for autonomous trading to become common before tightening controls. The practical starting point is to separate an agent’s capability from its authority. A model may be able to initiate a payment, change an account or place an order. That doesn't mean the company should give it permission to do so.
Hall calls for unique agent identities, narrow access, secure credentials, transaction limits, sandboxing and monitoring. Those controls let a firm identify which agent acted, restrict the systems it can reach, cap what it can do and revoke access when something goes wrong. Firms remain responsible for the agent’s actions.
OSFI’s July 2026 agentic AI guidance tells Canadian financial institutions to give agents unique identities, limit what they can access, require approval for higher risk actions and keep records of what each agent does. OSFI also wants firms to test failures and outages, understand which outside services support critical work and keep manual backups when those services fail.
AI Agents Expose Weak Links in Financial Authorization shows how the same problem plays out across financial systems. One task can pass through an AI model, cloud service, API, bank, custodian or payment system. Each system that can change money, ownership or customer instructions needs to know what the agent is allowed to do and who approved it.
Companies also need to know whether their AI suppliers are really independent. Two vendors can look separate while relying on the same model or cloud provider. A firm may have strong controls inside its own business and still depend on the same outside infrastructure as competitors and counterparties.
Talking Point
Financial firms should treat AI autonomy as an authorization decision, not a software feature. Start with the job the agent needs to perform, give it only the data and tools required, cap financial actions and require another approval before higher impact decisions change money, ownership or customer outcomes. Record what the agent does, test how it behaves when systems fail and keep a manual way to stop or complete critical work.
The same review should cover outside dependencies. Know which models, clouds and APIs support important workflows and where several suppliers depend on the same infrastructure. An agent can work exactly as designed inside one company and still add to market stress if many firms use similar technology and react the same way at once.
The National Crowdfunding & Fintech Association (NCFA Canada) is a financial innovation ecosystem that provides education, market intelligence, industry stewardship, networking and funding opportunities and services to thousands of community members and works closely with industry, government, partners and affiliates to create a vibrant and innovative fintech and funding industry in Canada. Decentralized and distributed, NCFA is engaged with global stakeholders and helps incubate projects and investment in fintech, alternative finance, crowdfunding, peer-to-peer finance, payments, digital assets and tokens, artificial intelligence, blockchain, cryptocurrency, regtech, and insurtech sectors. Join Canada's Fintech & Funding Community today FREE! Or become a contributing member and get perks. For more information, please visit: www.ncfacanada.org
 |  |  |
|---|---|---|
 |  |  |
Leave a Reply