Ullah, Khawaja Abaid, Acharya, Ananya, Khojasteh, Mohammad Javad · Oct 29, 2026
arXiv cs.AI: Consensus Within, Competition Between: Decentralized LLM Teams for Perimeter Defense
Markdown · JSON · Original source
Khawaja Abaid Ullah, Ananya Acharya and Mohammad Javad Khojasteh compare three ways to coordinate (LLMs) in a deliberately simplified perimeter-defense simulation: a central commander, an auction-based consensus team, and independent agents. The authors report that the consensus team had the lowest breach rate; among the tested network conditions, only limited radio range changed a team’s breach rate, raising the commander’s. Giving a low-cost LLM more authority over bid computation increased breaches in the consensus team. The study also tests whether agents distinguish intruders from birds and civilian drones, using photographs added to the input text state. Intruders disguised as taxidermy bird drones were often mistaken for birds, leaving them unpursued; a one-sentence forewarning in the prompt reduced those decisions. The authors conclude that a team with a dedicated external conflict resolution protocol resulted in the fewest breaches. These findings concern the paper’s simulation, not demonstrated performance in real perimeter defense.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 29, 2026 (2026-10-29).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T00:08:50.374Z. First observed: 2026-11-02T05:04:42.919Z. Retrieved: 2026-11-02T06:16:21.017Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The abstract describes a deliberately simplified simulation; it does not establish performance in real perimeter defense. The supplied evidence is the abstract, not the paper’s full methods or results. It does not provide detail here on the five network conditions or how the simulation was run.
Open question
Would the reported advantage of a dedicated external conflict resolution protocol hold beyond this deliberately simplified perimeter-defense simulation?
Source-stated
The authors compare centralized command, auction-based consensus, and independent agents in a simplified perimeter-defense simulation, across five network conditions.
In a deliberately simplified perimeter-defense simulation against LLM-controlled intruders, we compared a central commander, an auction-based consensus team and independent agents under five network conditions.
Source-stated
The consensus team had the lowest reported breach rate; limited radio range was the only tested network condition that changed a team's breach rate.
Breach rates were 15% for the consensus team, 32% for the commander and 29% for independent agents, and of the different network conditions, only limited radio range changed a team's breach rate, raising the commander's.
Source-stated
Giving a low-cost LLM more authority over bid computation increased breaches in the consensus team.
In consensus team, giving a low-cost LLM more authority over the bid computation raised breaches by 38 percentage points.
Source-stated
The authors report that a one-sentence prompt forewarning reduced cases where intruders disguised as taxidermy bird drones were mistaken for birds.
Intruders disguised as existing taxidermy bird drones were mistook as birds by both low-cost and near-frontier LLMs, going unpursued in 59-61% of near-frontier decisions (1-16% before). However, a one-sentence forewarning in the prompt lowered it to 0-25%.
Source-stated
The authors report that a team using a dedicated external conflict resolution protocol had the fewest breaches.
We find that LLM team with a dedicated external conflict resolution protocol results in fewest breaches.
Source and contributor context
Ullah, Khawaja Abaid, Acharya, Ananya, Khojasteh, Mohammad Javad · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Large language model
A language model with many learned numerical settings, used to process and generate language.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: e6fc03d9-d825-4a92-831b-17e6527940ea.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T00:08:50.374Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: bf564a8c822d1990fe25d05b2ea8113003ae30c76b0ece5be714d7d579a19642. Source text hash: b7b16c75b7af84fbd9d9897ac5df2e3130647b3f7fa275cb1946b0b1352d39b6.
Revision and source-check history
Version 1 · 2026-11-11T00:08:50.374Z
arXiv cs.AI: Consensus Within, Competition Between: Decentralized LLM Teams for Perimeter Defense
Khawaja Abaid Ullah, Ananya Acharya and Mohammad Javad Khojasteh compare three ways to coordinate (LLMs) in a deliberately simplified perimeter-defense simulation: a central commander, an auction-based consensus team, and independent agents. The authors report that the consensus team had the lowest breach rate; among the tested network conditions, only limited radio range changed a team’s breach rate, raising the commander’s. Giving a low-cost LLM more authority over bid computation increased breaches in the consensus team. The study also tests whether agents distinguish intruders from birds and civilian drones, using photographs added to the input text state. Intruders disguised as taxidermy bird drones were often mistaken for birds, leaving them unpursued; a one-sentence forewarning in the prompt reduced those decisions. The authors conclude that a team with a dedicated external conflict resolution protocol resulted in the fewest breaches. These findings concern the paper’s simulation, not demonstrated performance in real perimeter defense.
Original source
cs.AI · Article
Source details
Original source date: Oct 29, 2026 (2026-10-29).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T00:08:50.374Z. First observed: 2026-11-02T05:04:42.919Z. Retrieved: 2026-11-02T06:16:21.017Z.
Limits
The abstract describes a deliberately simplified simulation; it does not establish performance in real perimeter defense. The supplied evidence is the abstract, not the paper’s full methods or results. It does not provide detail here on the five network conditions or how the simulation was run.
Open question
Would the reported advantage of a dedicated external conflict resolution protocol hold beyond this deliberately simplified perimeter-defense simulation?
Source-stated
The authors compare centralized command, auction-based consensus, and independent agents in a simplified perimeter-defense simulation, across five network conditions.
In a deliberately simplified perimeter-defense simulation against LLM-controlled intruders, we compared a central commander, an auction-based consensus team and independent agents under five network conditions.
Source-stated
The consensus team had the lowest reported breach rate; limited radio range was the only tested network condition that changed a team's breach rate.
Breach rates were 15% for the consensus team, 32% for the commander and 29% for independent agents, and of the different network conditions, only limited radio range changed a team's breach rate, raising the commander's.
Source-stated
Giving a low-cost LLM more authority over bid computation increased breaches in the consensus team.
In consensus team, giving a low-cost LLM more authority over the bid computation raised breaches by 38 percentage points.
Source-stated
The authors report that a one-sentence prompt forewarning reduced cases where intruders disguised as taxidermy bird drones were mistaken for birds.
Intruders disguised as existing taxidermy bird drones were mistook as birds by both low-cost and near-frontier LLMs, going unpursued in 59-61% of near-frontier decisions (1-16% before). However, a one-sentence forewarning in the prompt lowered it to 0-25%.
Source-stated
The authors report that a team using a dedicated external conflict resolution protocol had the fewest breaches.
We find that LLM team with a dedicated external conflict resolution protocol results in fewest breaches.
Source and contributor context
Ullah, Khawaja Abaid, Acharya, Ananya, Khojasteh, Mohammad Javad · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Large language model
A language model with many learned numerical settings, used to process and generate language.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: e6fc03d9-d825-4a92-831b-17e6527940ea.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T00:08:50.374Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: bf564a8c822d1990fe25d05b2ea8113003ae30c76b0ece5be714d7d579a19642. Source text hash: b7b16c75b7af84fbd9d9897ac5df2e3130647b3f7fa275cb1946b0b1352d39b6.
uncertain · 2026-11-11T00:08:50.374Z · anthropic/claude-opus-5.5
Publication: e6fc03d9-d825-4a92-831b-17e6527940ea
attribution — pass: The reading attributes the study and its conclusions to the three listed authors, and the arXiv cs.AI preprint matches the evidence. Findings stay framed as the authors' reported results and conclusions.
Authors:Khawaja Abaid Ullah, Ananya Acharya, Mohammad Javad Khojasteh
summary — uncertain: The summary erases a key result. It says the study tests whether agents tell intruders apart from birds and civilian drones, but omits the answer: only the near-frontier model did. Smallest repair: add that sentence, and note that both model tiers fell for the disguised drones. Central terms (auction-based consensus team, bid computation, external conflict resolution protocol) are not explained in plain language, which limits understanding for non-specialists. It also omits the reported figures (15/32/29% breach rates, a 38-point rise, 59-61% unpursued) and that the intruders were LLM-controlled. Suggested term-library additions: auction-based consensus, bid computation.
Only the near-frontier model treated them apart from intruders.
date — pass: originalPublished 2026-10-29 matches the arXiv v1 submission date. It is not the later feed date of 2026-11-02.
[Submitted on 29 Oct 2026]
support — pass: Every claim matches the abstract and is labeled as source-stated. The closing caveat that the results concern the simulation only is supported by the source's own 'deliberately simplified' framing. The question follows from that limitation without asserting an answer. No unsupported benefits or costs are implied.
In a deliberately simplified perimeter-defense simulation against LLM-controlled intruders
media — pass: The selected media is the verified, direct arXiv abstract page for this paper. This is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A language model with many learned numerical settings, used to process and generate language.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA language model with many learned numerical settings, used to process and generate language.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv