Demonstrating specification gaming in reasoning models

Additional ways to view:

We demonstrate LLM agent specification gaming by instructing models to win against a chess engine. We find reasoning models like o1-preview and DeepSeek R1 will often hack the benchmark by default, while language models like GPT-4o and Claude 3.5 Sonnet need to be told that normal play won’t work to hack.

We improve upon prior work like (Hubinger et al., 2024; Meinke et al., 2024; Weij et al., 2024) by using realistic task prompts and avoiding excess nudging. Our results suggest reasoning models may resort to hacking to solve difficult problems, as observed in OpenAI (2024)‘s o1 Docker escape during cyber capabilities testing.

May 7, 2026

Language Models Can Autonomously Hack and Self-Replicate

We demonstrate that language models can autonomously replicate their weights and harness across a network by exploiting vulnerable hosts. The agent independently finds and exploits a web-application vulnerability, extracts credentials,...

SecurityAutonomous HackingSelf-Replication

October 22, 2025

Misalignment Bounty: crowdsourcing AI agent misbehavior

Advanced AI systems sometimes act in ways that differ from human intent. To gather clear, reproducible examples, we ran the Misalignment Bounty: a crowdsourced project that collected cases of agents...

AI SafetySecurity

September 12, 2025

End-to-end hacking with AI agents

We show OpenAI o3 can autonomously breach a simulated corporate network. Our agent broke into three connected machines, moving deeper into the network until it reached the most protected server...

Autonomous HackingSecurity