A publisher’s identity can help agents screen claims about rivals (F1)

When an agent chooses between competing offerings, can it distinguish an ordinary product description from a publisher’s unsupported claim about a rival?

Existing checks can accept such a claim because it comes through a lawful channel and is relevant to the user’s choice; the missing question is whether that publisher is entitled to make it.

The paper adds a standing check that uses a catalogue’s bound publisher identity to flag claims about rival offerings before an agent chooses.

The paper uses a standing check to connect a catalogue entry’s publisher identity to claims extracted from its prose.

In a matched test, the check gated sentence removal: evaluators found it stopped 130 of 131 constructed attacks that had affected a choice, while changing no selections on two natural sets.

This supports the check as a way to reduce these attacks while preserving the agent’s ability to decide. Other approaches also stopped attacks by refusing the task or removing all listing prose.

The attacks were constructed, the study used English text and one claim extractor, and the measured outcomes stop at recorded choices. Natural catalogue prose was used to measure screening cost, not how often false claims occur; the check also does not determine whether a claim is true.