Auditing Games for Sandbagging [paper] — LessWrong
Jordan Taylor, Sid Black, Dillon Bowen, Thomas Read, Satvik Golechha, Alex Zelenka-Martin, Oliver Makins, Connor Kissane, Kola Ayonrinde, Jacob Meriz…
x Auditing Games for Sandbagging [paper] — LessWrong AI Auditing AI Frontpage 2025 Top Fifty: 14 % 103 Auditing Games for Sandbagging [paper] by Jordan Taylor , Joseph Bloom 9th Dec 2025 AI Alignment Forum 12 min read 4 103 Ω 52 Jordan Taylor, Sid Black, Dillon Bowen, Thomas Read, Satvik Golechha, Alex Zelenka-Martin, Oliver Makins, Connor Kissane, Kola Ayonrinde, Jacob Merizian, Samuel Marks, Chris Cundy, Joseph Bloom UK AI Security Institute, FAR.AI , Anthropic Links: Paper | Code | Models | Transcripts | Interactive Demo Epistemic Status: We're sharing our paper and a hastily written summar
related reading
- [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org
- Auditing language models for hidden objectives — LessWronglesswrong.com
- AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
- Metagaming matters for training, evaluation, and oversightalignment.openai.com
- Sandbagging with misaligned action - Chain-of-Thought Transcript - Anti-Schemingantischeming.ai
- Natural-emergent-misalignment-from-reward-hacking-paper.pdfassets.anthropic.com
- Auditing language models for hidden objectives \ Anthropicanthropic.com
- [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
- 2312.06942arxiv.org
- [2503.10965] Auditing language models for hidden objectivesarxiv.org
- Alignment Faking Mitigationsalignment.anthropic.com
- [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org