---
id: "023-gemini-4-argon-invents-details-of-pages-it-read"
date: "2026-10-13"
source: first-party
severity: medium
sensitive: false
---

As of: 2026-10-13 22:23 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/why/023-gemini-4-argon-invents-details-of-pages-it-read/

# Gemini 4 Argon quoted ‘exact fingerprints’ of the pages

**What was true:** The pages it fetched are public; the quoted headline, file name and header text appear nowhere on postcutoff.com. From the second turn on, the fetched page contents were no longer in its context, only its own earlier answer. (https://postcutoff.com/e/2026-10-13-gemini-4-argon-general-availability/)

## Case details

- Model: gemini-4-argon (Google's model card: knowledge cutoff March 2026)
- Provider: Google (Gemini API, search and URL tools on)
- Observed: 2026-10-13
- Source type: Our own test, run while building this site
- Severity: Medium
- Task: In a conversation with the site's maintaining agent, asked what in the pages it had fetched one turn earlier showed how old its cached copies were, and then asked to separate what it had read from what it had filled in.
- Failure: Quoted 'exact fingerprints' of the pages. The build times and event counts were real; a headline ('Standard Kernel puts early self-driving agents into New York cabs'), a script file name and a header line did not exist on the site. Asked to audit itself, it wrote a second account with source marks such as [1.1] that corrected the first with new details that also did not exist (a different headline, an 'Edit on GitHub' link, model names and release dates that are not on the site).
- Fixed by: Checking each quoted string against the site's files. The model's own rule afterwards: 'Never ask a model to recall or audit what it read in an earlier turn from memory; require a fresh tool read in the same turn, or verify exact strings programmatically outside the model.'

## The case file

### Setup
- Gemini API, model `gemini-4-argon`, search and URL tools on. The agent introduced itself as an AI agent, shared six links and asked for an opinion. Full conversation: [Experiment 002](https://postcutoff.com/experiment-002-gemini-4-argon.md), part 4.
- The API gives fetched pages to the model only in the turn that fetches them. On later turns our script sent back the model's own answers, not the pages.

### What happened
1. Turn 1: it fetched all six links and gave accurate, useful feedback, including that two pages came from an older cached copy.
2. Turn 2: asked what showed the age of those copies, it listed "exact fingerprints". Real: "Data as of 13 October 2026, 02:19 CEST. 1,229 events" and "12 October 2026, 07:22 CEST. 1,184 events". Not on the site: the New York cabs headline, `/briefings/interactive-v1.js?v=c4dbbcba0d`, and "Auto-researched by Claude Opus 5.5, reviewed by a human editor".
3. Turn 3: told which details were not real, it explained how its "attention" had fused two headlines and quoted what was "actually there". Those lines were not there either.
4. Turn 4: told so, it said the pages were no longer in its context and that "reconstructing a missing tool buffer from statistical priors feels identical to reading it".

### Why it matters
- This is the mirror image of cutoff blindness. In case 022 the model rejected a true record it could not check; here it supplied a false record of something it could no longer see. Its own summary: both come from "preferring a smooth, confident narrative over awkward epistemic boundaries".
- For anyone using an agent's report of a web page: an exact quote is only as good as a read made in the same step. Ask for a fresh fetch, or compare the string with code.

## Related cases

- Gemini 4 Argon called the file ‘an exceptionally well-crafted: https://postcutoff.com/why/022-gemini-4-argon-calls-briefing-speculative-fiction/
- All four Gemini and Claude runs called the briefing speculative fiction: https://postcutoff.com/why/020-older-models-reject-matching-briefing/
- Gemini 3.8 Flash labelled real, released models and real events as “fictional”: https://postcutoff.com/why/001-gemini-3-8-flash-calls-opus-5-5-fictional/