mlsanityData to 14 Oct 2026

Recovering Agentic Sovereignty: Mitigating the Consensus Paradox via Contrastive Epistemic Decoding

arXiv 2609.25570 ↗First posted 22 Sep 2026cs.AIcited by 0

On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks also fits: Social simulation and multi-agent debate

Abstract (arXiv)

Large language models (LLMs) exhibit a parametric vulnerability to adversarial swarm consensus. To mitigate this sycophancy, we introduce Contrastive Epistemic Decoding (CED), a zero-shot inference intervention. Unlike standard Contrastive Decoding (CD) which relies on a weaker secondary model, CED utilizes a dual forward-pass on a single architecture to isolate conformity bias. By introducing a novel asymmetric, zero-bounded probability clamp and discrete top-k truncation mask, CED mathematically suppresses toxic consensus tokens without causing grammatical collapse. Evaluated across 7,200 paired trajectories on complex benchmarks (GAIA, SWE-bench, Multi-Challenge) using Gemma-2 (9B), Llama-3.1 (8B), and Mistral v0.3 (7B), CED successfully neutralizes architectural and positional biases. By reducing cognitive loafing by up to 33.00% absolute, CED drives significant performance gains, yielding up to a 30.75% accuracy recovery. Regaining sovereignty induces distinct architectural behaviors---passive task-focus in Gemma-2 and active refutation of the simulated swarm in Llama-3.1---showing CED decouples compliance from capability without fine-tuning.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 11 of its 19 references that are in the map. First posted on arXiv 22 Sep 2026.

Builds on

11 of 19 references.

Closest ancestors · 1

Same direction, newest first

  1. 2026Beyond the Attention Stability Boundary: Agentic Self-Synthesizing Reasoning Protocols27 Apr

Other references · 10

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  3. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,568
  4. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  5. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  6. 2023Improving Factuality and Reasoning in Language Models through Multiagent Debatecited by 1,784
  7. 2023Towards Understanding Sycophancy in Language Modelscited by 1,040
  8. 2023DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Modelscited by 376
  9. 2023Trusting Your Evidence: Hallucinate Less with Context-aware Decodingcited by 327
  10. 2026The Bystander Effect in Multi-Agent Reasoning: Quantifying Cognitive Loafing in Collaborative Interactionscited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Emergent Collaborative Deliberation in Multi-Model AI Systems: A BFT-Derived Protocol for Epistemic Synthesis
  2. 2026The Cognitive Penalty: Ablating System 1 and System 2 Reasoning in Edge-Native SLMs for Decentralized Consensus
  3. 2026The Consensus Trap: Rescuing Multi-Agent LLMs from Adversarial Majorities via Token-Level Collaboration
  4. 2026SFAD: Speculative Factuality-Aware Decoding
  5. 2026Representational Collapse in Multi-Agent LLM Committees: Measurement and Diversity-Aware Consensus
  6. 2026The Formalism Trap: Are LLM-as-a-Judge Evaluators Blinded by Consensus Mimicry under Social Load?
  7. 2026Resilient Consensus in Agentic AI
  8. 2026PEAR: Permutation-Equivariant Adaptive Routing Multi-Agent Debate