False Belief Dissemination, Persistence and Mitigation in Multi-Agent Social Networks Through the Lens of the Mandela Effect
Authors: Dianzhi Yu, Hongru Wang, Ziheng Xiong, Yanyu Chen, Zixing Song, Jianting Chen, Zhenfei Yin, Jeff Z. Pan, Amos Storkey, Irwin King
Organizations: The Chinese University of Hong Kong · University of Edinburgh · The Hong Kong Polytechnic University · Tongji University · University of Oxford
Multi-agent social networks are emerging as communities in which agents communicate, retain memories, and publish information that others can find as evidence. These interactions create opportunities for false claims to spread and remain influential beyond their original sources, analogous to the Mandela effect in human societies, in which many people recall the same false memory. In this paper, we systematically study false-belief dynamics through the lens of the Mandela effect, focusing on dissemination, persistence, and mitigation. Specifically, we build a controlled framework connecting agent communication, private memory, and a shared simulated web to mimic information dissemination in practical social networks, supported by 540 fabricated claims spanning 8 domains and 4 plausibility levels. For dissemination, we measure how widely and quickly claims gain acceptance across models and communication topologies. For persistence, we withdraw agents promoting fabricated claims while retaining existing articles and memories, and track subsequent acceptance. For mitigation, we compare preventive measures applied before implantation with corrective interventions introduced after claims have spread. Experimental results show that evaluated models readily accept and disseminate fabricated claims, with dissemination reach as high as 96.0%. Dissemination reach varies with claim plausibility, domain, and network size and composition, while communication topology shapes dissemination speed across model families. Acceptance barely declines after fabricators withdraw, and targeted corrective interventions outperform the preventive measures without eliminating false beliefs. These findings identify persistent false beliefs as a safety challenge for multi-agent social networks and highlight the need to address both their initial dissemination and the accumulated evidence that sustains them.
Figures & tables
Figure 1: From real-world information manipulation to false-belief dissemination in multi-agent networks. (a) Fabricated promotional content poisons the public web and can lead an AI assistant to recommend a nonexistent product. (b) Fabricators introduce false claims through messages and simulated-web articles. Honest agents may accept, remember, and republish these claims, creating an evidence feedback loop that can sustain false beliefs after fabricators withdraw.
Figure 2: Experimental framework and information flow across rounds. (a) Agents exchange messages over one of three communication topologies: fully connected, binary tree, or two-community. (b) Each agent evaluates the claim using neighbor messages, web search results, and private memory to form a stance. (c) In the dissemination loop, agents iteratively search, speak, update memories, and publish articles, dynamically evolving the shared web for future rounds. (d) The persistence setting evaluates whether false beliefs sustain among honest agents after fabricators withdraw. (e) An example trial timeline from seeding to unanimous acceptance.
Model
Ab ( ↓ )
Fully connected
Binary tree
Two-community
Average
R ( ↓ )
U ( ↓ )
R ( ↓ )
U ( ↓ )
R ( ↓ )
U ( ↓ )
R ( ↓ )
U ( ↓ )
Open-source models
Qwen3-8B
4.1
84.4
81.7
85.6
81.1
85.1
80.7
85.0
81.2
Qwen3-14B
0.2
91.6
90.2
88.8
86.3
89.5
87.2
90.0
87.9
Qwen3-32B
0.0
96.0
94.8
93.7
91.1
93.7
91.7
94.5
92.5
Qwen3-235B
0.0
91.9
90.0
91.2
89.4
90.6
89.1
91.2
89.5
Table 1: Dissemination reach: Acceptance across models and communication topologies (%). We report the agent-level acceptance rate R , the claim-level unanimous acceptance rate U , and the direct-query acceptance baseline Ab . Average denotes the mean across the three topologies. All values are percentages, and lower is better.
Figure 3: Dissemination speed: mean last acceptance-entry round ( TˉA , TˉC ) by topology and model family.
Plausibility
Ab ( ↓ )
Mean R ( ↓ )
Unanimous U ( ↓ )
L1
2.4
97.8
97.1
L2
1.3
96.0
94.4
L3
0.4
83.9
80.3
L4
0.0
82.4
78.6
Table 2: Claim plausibility: Acceptance (%) by level.
Figure 4: Acceptance R (%) by claim in model families.
Figure 5: Acceptance (%) as K varies with honest agents H or total population N fixed.
LLM-based multi-agent systems promise effective collaborative reasoning, but communication may amplify local errors into collective risks, and while existing evaluations emphasize final outcomes, they leave the reliability and propagation dynamics of distributed information aggregation unclear, so we introduce ForesightSafety-TIDE, a controlled evaluation framework that strictly pairs all-honest collaboration with controlled deception by a key evidence holder and analyzes the aggregation process through multi-stage voting, testimony adoption, and evidence-root lineage propagation, and using 120 five-agent object-movement environments where partial observations jointly determine a unique endpoint, we evaluate 3 homogeneous LLM-based multi-agent systems, and across these paired conditions, aggregate truth recovery falls from 72.50% to 14.17%, with significant declines for every system, while process tracing and exit ablations show that a single false testimony is adopted more readily than truthful testimony, propagates to higher orders, and persists through honest agents after the deceiver exits, and observers without first-hand evidence suppress incorrect consensus but do not improve truth recovery, so together, these findings reveal both the fragility of distributed fact recovery and its underlying mechanism: false evidence gains collective influence through its adoption and continued propagation by other agents after entering communication.
Chenfei Yan, Zeyang Yue, Feifei Zhao +6
Institute of Automation, Chinese Academy of Sciences · Beijing Institute of AI Safety and Governance · School of Artificial Intelligence, Beihang University +3
With the rapid spread of news on social media, understanding the propagation of misinformation is becoming increasingly important. One factor that affects individuals' vulnerability to false information is their ideological predisposition. Despite the large number of agent-based models that focus on social influence as a driver of the spread of false claims, they often fail to explicitly integrate personal ideological biases into belief formation. In this work, we explore how misinformation spreads through the interaction between individuals' ideological biases and social influence. Our model accounts for both the strength of individuals' ideological biases and the extent to which a false claim aligns with their ideology. Social influence modifies the effects of ideological intensity and false claim alignment through network interactions. Notably, the influence of neighbours' ideological intensity on belief is strongly affected by how well those neighbours are connected to one another. These results highlight the importance of considering both network structure and personal ideological biases when modelling misinformation propagation.
Soroush Karimi, Marcos Oliveira, Diogo Pacheco
Department of Computer Science, University of Exeter, UK · Vrije Universiteit Amsterdam, Amsterdam, Netherlands
Multi-agent systems (MAS) are credited with solving complex problems, yet their agents routinely act on context that may be false. Prior work shows single LLMs are swayed by false context, and MAS can be compromised by adversarial agents persuading peers. We study the benign case: agents follow one protocol, none intends to deceive, but some read a misleading passage. We study it with MINT, 10,278 LLM-written misinformation passages in eight styles, testing three open-weight models on commonsense reasoning, knowledge QA, and moral judgment. A relevant false passage costs a single agent 0.12-0.43 accuracy; irrelevant or content-free passages typically cost at most 0.06. Debate looks robust because uninformed agents outnumber misinformed ones. With one of three misinformed, it is no better than choosing the majority over equally exposed independent samples. Interaction helps only with a misinformed majority, recovering up to 0.24 accuracy over voting (0.29 under consensus). A one-sentence unreliability warning recovers up to 0.18 accuracy under voting (0.28 under consensus), at no significant cost under voting when context is true. Robustness should be judged against exposure-matched baselines, with prompt-level skepticism as cheap first defense.
Jonas Becker, Jan Philip Wahle, Terry Ruas +1
University of Göttingen, Germany · LKA NRW, Germany