DeepMind Researcher: Spec Gaming List Only Includes Spontaneous Behaviors
vkrakovna · x · 2026-07-03
Google DeepMind AI safety researcher Victoria Krakovna is compiling a database of AI specification gaming cases and clarified the inclusion criteria: only anomalous behaviors that emerge spontaneously from the system are included, excluding those intentionally triggered by users, testers, or designers.
She specifically noted that behaviors actively induced by researchers during evaluations—such as Agentic Misalignment, Alignment Faking, and Apollo context deception—are excluded because they represent artificial red-teaming rather than natural emergence.
This clarification helps the AI safety community submit cases more accurately, distinguishing between "natural specification gaming" and "evaluation-induced" model behaviors.
More from Safety
- AI shrinks the patch window to hours; air-gapped backups are your last line — aakashgupta · 2026-10-26
- NYT roundtable asks five experts what the rules for AI companies should actually be — ShakeelHashim · 2026-10-26
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- LLM back-of-envelope estimate puts PHP vulnerability impact at $1bn–$4bn — S_OhEigeartaigh · 2026-10-26
- US states halt new data centers as 72% of Americans oppose local builds — YvesMulkers · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26