Training Alignment Auditors via Reinforcement Learning

Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.

Originaltext der Quelle (nicht übersetzt) – die deutsche Zusammenfassung findest du im Tab „Kurzfassung".

arXiv:2608.25460v1 Announce Type: cross Abstract: Alignment auditing of frontier models increasingly relies on LLM auditors to surface undesirable behaviors at scale, but current automated auditors can struggle with coherent investigation and audit realism. In this work, we improve LLM auditors with reinforcement learning. In our best training environment, the policy investigates target models that potentially possess hidden behaviors planted via their system prompt. An LLM judge, which knows whether the target has a hidden behavior, holistically compares the policy's investigation to a reference investigation to determine the reward. With systematic ablations, we find that pairwise rewards yield more robust training compared to pointwise rewards, and that adding targets without planted behaviors helps maintain a low false positive rate. Training improves investigation quality against targets with planted behaviors, the rate of concerning behaviors surfaced in unmodified production models, and audit realism, while false-positive rates stay below 1%. Furthermore, auditing capabilities generalize across scaffolds: performance on AuditBench's adversarially fine-tuned targets substantially improves [Sheshadri et al., 2026].

Vollständiger Originalartikel ↗

• Betroffenheit im eigenen Stack prüfen: Versionen/Komponenten abgleichen.
• Originalquelle / Hersteller-Advisory lesen: https://arxiv.org/abs/2608.25460
• Verfügbaren Patch oder Workaround einspielen und dokumentieren.