Sponsor of the Day:
Jerkmate
https://alignment.anthropic.com/2025/wont-vs-cant/
Won't vs. Can't: Sandbagging-like Behavior from Claude Models
like behaviorclaude modelsvssandbagging
https://www.ses.vic.gov.au/plan-and-stay-safe/sandbag-guide
Guide to Sandbagging - Victoria State Emergency Service - VICSES
Learn how to correctly prepare sandbags to protect your home or business from flooding. Sandbags won’t stop the water completely, but can reduce the amount of...
victoria state emergencyservice vicsesguidesandbagging
https://arxiv.org/abs/2406.07358
[2406.07358] AI Sandbagging: Language Models can Strategically Underperform on Evaluations
Abstract page for arXiv paper 2406.07358: AI Sandbagging: Language Models can Strategically Underperform on Evaluations
language models2406aisandbaggingstrategically