Adaptive LLM-Driven Honeypots vs. LLM-Driven Attackers: Does Narrative Coherence Improve Deception?

According to the abstract, Pedro Beltrán-López and colleagues test whether a language-model honeypot that builds a coherent story from linked fake files, processes and logs keeps an autonomous LLM attacker engaged longer. They ran 1,150 sessions comparing four conditions of increasing sophistication.

Why it matters

The same thing, for different readers.

For lawmakers

As autonomous AI attackers become realistic, defenders are testing AI-driven decoys; officials weighing cyber defence may want evidence on whether such deception works.

For researchers

The study pits an AI defender against an AI attacker in live SSH sessions, offering a controlled setup for measuring deception against autonomous offensive agents.

How the finding was reached

In our words, from the source. Weigh the finding by how it was made.

What was done
An LLM defender and an autonomous LLM attacker interact over a real SSH session on an emulated Ubuntu host; 1,150 sessions compare four deception conditions across four attacker and four defender models.
What it found
The excerpt states the design and question but cuts off before the results, so no finding is given here.
What it does not show
Read from a truncated abstract excerpt; results were not seen.
Status
A preprint: not yet peer reviewed

The record

Questions
How is AI used in attacks, defence and war?
People
Pedro Beltrán-López (Wrote it)
What we read
We have read only the description the publisher gives. No quotations until someone reads the source itself.
Found by
The Radar, run of November 18, 2026, via arxiv-ai-safety-query (tier 1: the source’s own channel or an official record)
Cite this
arXiv, “Adaptive LLM-Driven Honeypots vs. LLM-Driven Attackers: Does Narrative Coherence Improve Deception?”, November 17, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/arxiv-2026-11-17-llm-honeypots-narrative-deception
Our notesLight lane: written from the triage excerpt only; the excerpt ends before the results.