2410.21514
arxiv.org · 8,141 words · saved by 1 readers
N/A
Sabotage Evaluations for Frontier Models Joe Benton, Misha Wagner, Eric Christiansen, Cem Anil, Ethan Perez, Jai Srivastav, Esin Durmus, Deep Ganguli, Shauna Kravec, Buck Shlegeris*, Jared Kaplan, Holden Karnofsky† , Evan Hubinger, Roger Grosse, arXiv:2410.21514v1 [cs.LG] 28 Oct 2024 Samuel R. Bowman, David Duvenaud…
related reading
- [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
- 2312.06942arxiv.org
- Agentic misalignment: How LLMs could be insider threats \ Anthropicanthropic.com
- Pre-deployment auditing can catch an overt saboteuralignment.anthropic.com
- AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
- [2502.17424] Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMsarxiv.org
- An alignment assessment of recent cybersecurity incidentsanthropic.com
- Agentic Misalignment in Summer 2026alignment.anthropic.com
- Natural-emergent-misalignment-from-reward-hacking-paper.pdfassets.anthropic.com
- [2401.05566] Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingarxiv.org
- Sabotage evaluations for frontier models \ Anthropicanthropic.com
- Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face — LessWronglesswrong.com