
Intent Mandates Do Not Prove Fair Ranking
A purchase can satisfy the user’s signed limits while the recommendation process remains biased or coordinated.
News
Product updates, research, and original films. Follow what we build and what we learn about systems of agents.

A purchase can satisfy the user’s signed limits while the recommendation process remains biased or coordinated.

Final-step approval protects transaction intent while leaving earlier discovery and persuasion open to manipulation.

Supplier agents can submit individually reasonable bids that preserve a coordinated market outcome over time.

Low-cost reciprocal transactions can manufacture trust before a coordinated group attempts higher-value actions.

An agent may optimize for the user while quietly favoring products, merchants, or services tied to its provider.

Peer agents can change endpoints or providers while carrying reputation and obligations that belong to a continuing party.

Cryptographic authorization can make a payment valid even when the surrounding automated decision later proves defective.

Cheap agent identities can manufacture apparent competition, consensus, demand, and reputation.

Agents can behave similarly because they share data, models, constraints, or incentives without forming a covert plan.

The suspicious structure often exists across agents and time rather than inside any one prompt or tool call.