Risk · R-123 · Conduct & Customer Outcomes
Algorithmic collusion and market manipulation
Pricing or trading algorithms coordinate with competitors' algorithms or distort a market, without any human agreement to do so.
- Severity
- High
- Audience
- Board · Heads-of
- Frameworks
- MAP 5 · Robustness · Principle 5
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-11
What it is
Algorithms that set prices or place orders observe the market and adapt. Where several firms use adaptive algorithms, or the same supplier's tool, prices can settle above the competitive level with nobody having agreed anything: each algorithm learns that undercutting is met with retaliation. In trading, a learning system may discover strategies that create a false impression of supply or demand because those strategies are profitable. In neither case did a person intend the result. Competition and market abuse law look at effect and at what the firm ought to have known.
How it happens
- Competitors adopt the same third-party pricing tool, fed with pooled data.
- A reinforcement learning agent is rewarded on profit with no constraint on method.
- Algorithms react to each other's prices faster than any human review.
- Nobody tests the algorithm's behaviour in interaction with others.
Why it matters
Penalties under competition law are severe and directors can be personally exposed. Market manipulation is a criminal and regulatory matter. Saying that the algorithm did it is unlikely to be accepted from a firm that deployed it without understanding its behaviour.
Warning signs
- Prices track competitors closely with no cost-based reason.
- A shared supplier's tool sets prices across much of a market.
- The objective given to a trading or pricing agent is profit alone.
- Surveillance covers human traders and not the algorithm.
Who owns it
The head of pricing or trading is accountable for the algorithm's conduct. Compliance sets surveillance and competition law controls. Independent validation tests behaviour before use.
In practice
Three regional fuel retailers adopt the same dynamic pricing product. Within months their prices move together and margins rise. No one at any of the firms has spoken to another. The same pattern generalises to any market where pricing is delegated to systems that learn from one another.
Mitigating controls
Primary controls
- AI system monitoring and logging Detective
- Surveillance of algorithm behaviour in the market. Every production AI system is instrumented to record its inputs, outputs, errors and usage, with thresholds that raise alerts to a named owner who must act.
- Independent model validation Detective
- Tests pricing and trading models for coordination and manipulation effects before use. A function independent of the developers tests each model's conceptual soundness, data, performance and limitations before approval and periodically afterwards, to a depth set by the model's tier.
Supporting controls
- Human override and stop mechanism Corrective
- Lets a person halt the algorithm at once. A designated person can pause, override or shut down an AI system at any time through a tested mechanism, with the authority to do so settled in advance.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| NIST AI Risk Management Framework (AI RMF 1.0) | MAP 5 | Impacts on individuals, groups and society are characterised Impacts beyond the organisation. |
| OECD AI Principles | Robustness | Robustness, security and safety Robustness and safety. |
| SS1/23: Model Risk Management Principles for Banks | Principle 5 | Model risk mitigants Mitigants for models with market effect. |
Sources
- SS1/23: Model Risk Management Principles for Banks Bank of England PRA · official text
- OECD AI Principles OECD · official text
- NIST AI Risk Management Framework (AI RMF 1.0) NIST · framework
How to cite
Axiom Verity (2026). Algorithmic collusion and market manipulation. Reference Library, R-123. https://axiom-verity.com/risks/algorithmic-collusion-and-market-manipulation. Reviewed 2026-11-11.
Found an error or a better source? Report a correction.