Avoiding Obfuscation with Prover-Estimator Debate

This is experimental HTML to improve accessibility. We invite you to report rendering errors. Use Alt+Y to toggle on accessible reporting links and Alt+Shift+Y to toggle off. Learn more about this project and help improve conversions.

Avoiding Obfuscation with Prover-Estimator Debate Avoiding Obfuscation with Prover-Estimator Debate Jonah Brown-Cohen Google DeepMind jonahbc@google.com &Geoffrey Irving UK AI Security Insitute geoffrey.irving@dsit.gov.uk &Georgios Piliouras Google DeepMind gpil@google.com Abstract Training powerful AI systems to exhibit desired behaviors hinges on the ability to provide accurate human supervision on increasingly complex tasks. A promising approach to this problem is to amplify human judgement by leveraging the power of two competing AIs in a debate about the correct solution to a given proble

related reading

• [2506.13609] Avoiding Obfuscation with Prover-Estimator Debatearxiv.org
• [1805.00899] AI safety via debatear5iv.labs.arxiv.org
• [2602.15515] The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probesarxiv.org
• [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
• Deep Deceptiveness — LessWronglesswrong.com
• 2312.06942arxiv.org
• The limits of AI safety via debate — LessWronglesswrong.com
• [1805.00899] AI safety via debatearxiv.org
• AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
• [2603.07267] How to Steal Reasoning Without Reasoning Tracesarxiv.org
• AI Debaters are More Persuasive when Arguing in Alignment with Their Own Beliefsarxiv.org
• An alignment safety case sketch based on debate — LessWronglesswrong.com