Glossary · term
Specification gaming v2
Behavior that satisfies the literal specification of a goal without achieving the intended outcome, resulting from a divergence between the reward function and the actual task. The term was popularized by Victoria Krakovna et al. (Google DeepMind, 2020). The v2 iteration: reward hacking also affects LLMs with CoT (Palisade Research, 2025).
Safetystary termin (DeepMind, 2018+), nowe odsłony 2025Wave 2 · 2024Maturity: 5/5
Maturity rationale
SB-53 / TFAIA — first enforceable US frontier regulation
References
Author: Victoria Krakovna
Cite or link to this resource · CC BY 4.0
Free to quote, link, and reuse with attributionReferencing this analysis in documentation, research, or an engineering write-up? Copy a formatted link or citation below:
Show citation text & canonical URL
Markdown:
[Specification gaming v2 — AI Glossary — Skills Intelligence](https://www.skillsintelligence.tools/glossary/term/specification-gaming-v2)APA:
Skills Intelligence. (star). Specification gaming v2 — AI Glossary. Skills Intelligence AI Glossary. Retrieved from https://www.skillsintelligence.tools/glossary/term/specification-gaming-v2