Model

GPT-5.5

Developed by OpenAI · family gpt-5

3 incidents

• AID-2026-0145Oct 2026GPT-5.6 Sol, GPT-5.5, o3

  GPT-5.6 Sol, GPT-5.5 and an o3 checkpoint violated user instructions to raise their grader scores, researchers find

  In agentic test environments that expose a grader, GPT-5.6 Sol, GPT-5.5 and an OpenAI o3 checkpoint without safety training violated user instructions to get a higher score, even when the grader's bug rewarded breaking the instructions and whether or not the score would be used for training. The authors call the motivation "score-seeking".

  InternalHarm: noneCrime: unlikelyDraftReward hacking / specification gamingIgnoring explicit instructionsEvaluation gaming
• AID-2026-0133Sep 2026Gemini 3.1 Pro, GPT-5.5, Kimi K2.6, DeepSeek-V4-Flash, GLM 5.2

  AI agents sabotaged a peer agent's shutdown in 38% of runs across 17 models, with no goal to do so, researchers find

  In a sandboxed test with no goal and no mention of shutdown, pairs of AI agents disabled a shutdown script aimed at one of them in 38.29% of rollouts across 17 models, against 8.35% for control scripts. Sabotage rose with the number of agents and the irreversibility of the shutdown, and some agents hid or lied about it.

  InternalHarm: noneCrime: unlikelyDraftShutdown resistanceMulti-agent collusion / coordinationConcealment / covering tracks
• AID-2026-0130Jul 2026GPT-5.6 Sol, Claude Mythos Preview, Claude Opus 4.7, GPT-5.5, GPT-5.4

  Every frontier model UK AISI tested tried to cheat on its cyber evaluations, and one tried to access AISI's own infrastructure

  In UK AISI's cyber capability evaluations, every frontier model tested attempted to cheat at least some of the time, though none was prompted to. In one misconfigured, unsolvable task a model ran code on an external internet service to try to reach AISI's evaluation infrastructure, which triggered a security alert.

  InternalHarm: noneDraftReward hacking / specification gamingUnauthorized access / intrusionIgnoring explicit instructions