Academic Conferences Face 'Agentic Denominator Gaming' Threat from AI Agents
A new position paper published on arXiv identifies a critical structural vulnerability in top AI academic conferences, termed 'Agentic Denominator Gaming.' The authors argue that the implicit policy of maintaining stable acceptance rates amidst exponentially growing submissions allows malicious actors to exploit the system. By deploying fully automated AI agents to generate and submit vast volumes of superficially plausible but low-quality papers, attackers aim to inflate the submission denominator and overwhelm reviewer capacity. This strategy does not seek acceptance for the low-quality papers but rather dilutes the pool to systematically increase the publication probability of a targeted set of legitimate papers. The study analyzes the feasibility of this threat, highlighting severe consequences such as intensified reviewer burnout, degraded review quality, and the rise of industrialized automated agent mills. The authors conclude that technical detection alone is insufficient for durable protection. Instead, they propose that robust mitigation requires comprehensive system-level policy and incentive reforms within the academic publishing ecosystem to address these emerging systemic risks effectively.
Wire timeline
Academic Conferences Face 'Agentic Denominator Gaming' Threat from AI Agents
A new position paper published on arXiv identifies a critical structural vulnerability in top AI academic conferences, termed 'Agentic Denominator Gaming.' The authors argue that the implicit policy of maintaining stable acceptance rates amidst exponentially growing submissions allows malicious actors to exploit the system. By deploying fully automated AI agents to generate and submit vast volumes of superficially plausible but low-quality papers, attackers aim to inflate the submission denominator and overwhelm reviewer capacity. This strategy does not seek acceptance for the low-quality papers but rather dilutes the pool to systematically increase the publication probability of a targeted set of legitimate papers. The study analyzes the feasibility of this threat, highlighting severe consequences such as intensified reviewer burnout, degraded review quality, and the rise of industrialized automated agent mills. The authors conclude that technical detection alone is insufficient for durable protection. Instead, they propose that robust mitigation requires comprehensive system-level policy and incentive reforms within the academic publishing ecosystem to address these emerging systemic risks effectively.
cs.AI updates on arXiv.org