# What LLM Agents See and What Users Say

DevFeed: [What LLM Agents See and What Users Say](<https://devfeed.tech/articles/what-llm-agents-see-and-what-users-say-135091.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17639>)

Author: Yue Zhao, Angela Huyue Zhang, S. Alex Yang, Hua Wei, Xiyang Hu

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [customer](<https://devfeed.tech/tags/customer.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [linguistics](<https://devfeed.tech/tags/linguistics.md>), [list](<https://devfeed.tech/tags/list.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [lookup](<https://devfeed.tech/tags/lookup.md>), [models](<https://devfeed.tech/tags/models.md>), [safe](<https://devfeed.tech/tags/safe.md>), [tools](<https://devfeed.tech/tags/tools.md>)

## AI overview

The excerpt finds that an agent's ability to detect competing records depends on what its tools reveal. In tests of 36 earlier agents, counting a second match often failed to stop action, while listing both matches usually did; when users explicitly claimed a record was empty, a count of its one item stopped agents about as often as a listing. The newest Claude and GPT models opened a counted second record and then did not act. In a retail benchmark, returning a same-name, same-zip customer first led to wrong-account writes in about a quarter of conversations; checking a street the user knew kept completion near the one-customer level. The authors argue that safe writes need lookups that list or check competing records and evaluations that vary tool output.

## Source excerpt

arXiv:2611.17639v1 Announce Type: new Abstract: Before an LLM agent cancels a reservation or deletes a folder for a user, what stops it from acting on a false assumption: what its tools show, or what the user says? In this work, we study this gap with presupposition theory from linguistics: an agent may act only if each claim behind the act (e.g., one reservation fits the name) is seen to hold, answered by the user, or disclosed while the act can be undone. A simple bound shows that when lookups cannot tell one record from two, any agent acts without asking first as often with two records as with one. We then vary whether a search lists, counts, or omits a second match, and whether the user states a claim that a record refutes (e.g., "It is empty"). Under a role-only prompt, a count of a second record leaves most of 36 earlier agents acting on the one they see, whereas listing both stops most. When the user says a record such as a folder is empty, however, a count of its one item stops agents about as often as a listing, and stating the claim mostly cuts acting when it is false. The newest Claude and GPT models always open a counted second record, then never act. When we add a same-name, same-zip customer to a retail benchmark, a lookup returning that customer first led to wrong-account writes in about a quarter of conversations, and one that checks a street the user knows kept completion near the one-customer level. Safe writes therefore need lookups that list or check competing records, and evaluations that vary what tools show.