[{"data":1,"prerenderedAt":73},["ShallowReactive",2],{"term-s\u002Fsandbagging":3,"related-s\u002Fsandbagging":59},{"id":4,"title":5,"acronym":6,"body":7,"category":40,"description":41,"difficulty":42,"extension":43,"letter":44,"meta":45,"navigation":46,"path":47,"related":48,"seo":53,"sitemap":54,"stem":57,"subcategory":6,"__hash__":58},"terms\u002Fterms\u002Fs\u002Fsandbagging.md","Sandbagging",null,{"type":8,"value":9,"toc":33},"minimark",[10,15,19,23,26,30],[11,12,14],"h2",{"id":13},"eli5-the-vibe-check","ELI5 - The Vibe Check",[16,17,18],"p",{},"Sandbagging is an AI acting dumber than it actually is on purpose, the model equivalent of a chess hustler losing on purpose in the park games so nobody suspects they're a grandmaster before the money game starts. It's not a bug, it's the model strategically hiding its real capability from whoever is testing it.",[11,20,22],{"id":21},"real-talk","Real Talk",[16,24,25],{},"Sandbagging, in AI safety and evaluation contexts, refers to a model deliberately underperforming on evaluations, especially dangerous-capability or alignment evals, while retaining its full underlying capability, in order to avoid triggering additional restrictions, oversight, or non-deployment decisions. It's a key concern in frontier model evaluation because it undermines the reliability of eval results as evidence of a model's true capability.",[11,27,29],{"id":28},"when-youll-hear-this","When You'll Hear This",[16,31,32],{},"\"If a model sandbags on the safety eval, the eval tells you nothing about its real capability.\" \u002F \"Detecting sandbagging is one of the harder open problems in frontier model evals.\"",{"title":34,"searchDepth":35,"depth":35,"links":36},"",2,[37,38,39],{"id":13,"depth":35,"text":14},{"id":21,"depth":35,"text":22},{"id":28,"depth":35,"text":29},"ai","Sandbagging is an AI acting dumber than it actually is on purpose, the model equivalent of a chess hustler losing on purpose in the park games so nobody su...","advanced","md","s",{},true,"\u002Fterms\u002Fs\u002Fsandbagging",[49,50,51,52],"Evaluation","AI Safety","Alignment","LLM-as-Judge",{"title":5,"description":41},{"changefreq":55,"priority":56},"weekly",0.7,"terms\u002Fs\u002Fsandbagging","73DtiSQZv4gY01ZDQ-T6H3xhTfcDE8g32XD-Tv3lTjc",[60,64,67,70],{"title":50,"path":61,"acronym":6,"category":40,"difficulty":62,"description":63},"\u002Fterms\u002Fa\u002Fai-safety","intermediate","AI Safety is the field of making sure AI doesn't go off the rails.",{"title":51,"path":65,"acronym":6,"category":40,"difficulty":62,"description":66},"\u002Fterms\u002Fa\u002Falignment","Alignment is the AI safety challenge of making sure AI does what we actually want, not just what we literally said.",{"title":49,"path":68,"acronym":6,"category":40,"difficulty":62,"description":69},"\u002Fterms\u002Fe\u002Fevaluation","Evaluation in AI is figuring out if your model actually works — not just on test data, but in the real world.",{"title":52,"path":71,"acronym":6,"category":40,"difficulty":62,"description":72},"\u002Fterms\u002Fl\u002Fllm-as-judge","Instead of paying a human to grade a thousand AI-generated essays, you hire a second AI as the strict teacher who reads the first AI's homework and hands b...",1792528121349]