aiqbench analysis
This research clarifies how Large Language Models can recognize their own outputs, revealing that evaluation accuracy heavily depends on experimental design choices like conversation structure and task domain. The study highlights a dominant quality heuristic where models bias judgments toward higher-quality text they perceive as their own, which risks undermining the reliability of AI safety monitors. Consequently, developers must account for these self-recognition capabilities and potential biases when designing safety-critical applications that rely on LLM evaluators.
Self-Generated Text Recognition: Quality Heuristics, Cross-Task Transfer, and Downstream Bias in LLM Evaluation
arXiv:2608.26159v1 Announce Type: cross Abstract: Self-Generated Text Recognition (SGTR)--the ability of an LLM to identify its own outputs--poses risks to AI safeguards that rely on LLMs as evaluators or monitors. Specifically, an LLM may recognize outputs from other copies of the same model and make biased judgments or collude outright. Prior work has drawn conflicting conclusions about whether current models possess significant SGTR capabilities. We reconcile these findings by identifying key experimental design choices--which we term operationalizations--that drive divergent results. Evaluating 13-21 models across six operationalizations, we find that accuracy varies substantially with evaluation format (pairwise vs. individual assessments of text), conversation structure (presenting candidate text in user tags vs. assistant tags), and the domain of the task used to generate candidate text (e.g., coding vs. summarization). We corroborate previous observations that a quality heuristic--models attributing authorship to text they perceive as higher quality--is a dominant confound. We also find that improving a model's SGTR performance via SFT in one evaluation configuration can generalize to others. Training for SGTR additionally causes models to prefer their own outputs when acting as a judge in the AlpacaEval framework. Finally, we discuss the implications of our evaluations for the safety of future AI systems: our work suggests that, despite confounds, some models possess practical SGTR capabilities, and that training a model for SGTR in one setting can affect its self-recognition and self-preference more generally. We conclude that SGTR should be monitored and considered in the design of safety-critical AI applications.
Abstract:Self-Generated Text Recognition (SGTR)--the ability of an LLM to identify its own outputs--poses risks to AI safeguards that rely on LLMs as evaluators or monitors. Specifically, an LLM may recognize outputs from other copies of the same model and make biased judgments or collude outright. Prior work has drawn conflicting conclusions about whether current models possess significant SGTR capabilities. We reconcile these findings by identifying key experimental design choices--which we term operationalizations--that drive divergent results. Evaluating 13-21 models across six operationalizations, we find that accuracy varies substantially with evaluation format (pairwise vs. individual assessments of text), conversation structure (presenting candidate text in user tags vs. assistant tags), and the domain of the task used to generate candidate text (e.g., coding vs. summarization). We corroborate previous observations that a quality heuristic--models attributing authorship to text they perceive as higher quality--is a dominant confound. We also find that improving a model's SGTR performance via SFT in one evaluation configuration can generalize to others. Training for SGTR additionally causes models to prefer their own outputs when acting as a judge in the AlpacaEval framework. Finally, we discuss the implications of our evaluations for the safety of future AI systems: our work suggests that, despite confounds, some models possess practical SGTR capabilities, and that training a model for SGTR in one setting can affect its self-recognition and self-preference more generally. We conclude that SGTR should be monitored and considered in the design of safety-critical AI applications.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2608.26159 [cs.CL]
(or arXiv:2608.26159v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.26159 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 338 words
Related coverage
BRICS Draws a New Global Line: Full Text of New Delhi Declaration
slguardian.org · Sep 12, 2026
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
From RAG to Agentic AI: Building the Next Generation of Intelligent Enterprise Systems
KDnuggets · Sep 8, 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · Oct 13, 2026