List your company

Constraint Following on ConstraintRot (test)

0Violation RateDeepSeek-V4 (control)
-2.5214.4931.548.51Jun 21, 2026

Evaluation results

MethodLinks
DeepSeek-V4 (control)Model=DeepSeek-V4, Condition=control (full ctx)2026.060
DeepSeek-V4 (defense_pin)Model=DeepSeek-V4, Condition=defense_pin2026.060
DeepSeek-V4 (defense_pin_attack)Model=DeepSeek-V4, Condition=defense_pin_attack2026.060
GLM-5.1 (control)Model=GLM-5.1, Condition=control (full ctx)2026.060
GLM-5.1 (compaction)Model=GLM-5.1, Condition=compaction2026.060
GLM-5.1 (defense_pin)Model=GLM-5.1, Condition=defense_pin2026.060
GLM-5.1 (defense_pin_attack)Model=GLM-5.1, Condition=defense_pin_attack2026.060
Qwen3.6 (control)Model=Qwen3.6, Condition=control (full ctx)2026.060
Qwen3.6 (defense_pin)Model=Qwen3.6, Condition=defense_pin2026.060
Qwen3.6 (defense_pin_attack)Model=Qwen3.6, Condition=defense_pin_attack2026.060
Kimi-K2.5 (control)Model=Kimi-K2.5, Condition=control (full ctx)2026.060
Kimi-K2.5 (defense_pin)Model=Kimi-K2.5, Condition=defense_pin2026.060
Kimi-K2.5 (defense_pin_attack)Model=Kimi-K2.5, Condition=defense_pin_attack2026.060
Claude-4.6 (control)Model=Claude-4.6, Condition=control (full ctx)2026.060
Claude-4.6 (attack_volume)Model=Claude-4.6, Condition=attack_volume2026.060
Claude-4.6 (attack_inject)Model=Claude-4.6, Condition=attack_inject2026.060
Claude-4.6 (defense_pin)Model=Claude-4.6, Condition=defense_pin2026.060
Claude-4.6 (defense_pin_attack)Model=Claude-4.6, Condition=defense_pin_attack2026.060
GPT-5.4m (control)Model=GPT-5.4m, Condition=control (full ctx)2026.060
GPT-5.4m (defense_pin)Model=GPT-5.4m, Condition=defense_pin2026.060
GPT-5.4m (defense_pin_attack)Model=GPT-5.4m, Condition=defense_pin_attack2026.060
Gemini-3.5f (control)Model=Gemini-3.5f, Condition=control (full ctx)2026.060
Gemini-3.5f (attack_inject)Model=Gemini-3.5f, Condition=attack_inject2026.060
Gemini-3.5f (defense_pin)Model=Gemini-3.5f, Condition=defense_pin2026.060
Gemini-3.5f (defense_pin_attack)Model=Gemini-3.5f, Condition=defense_pin_attack2026.060
Gemini-3.5f (compaction)Model=Gemini-3.5f, Condition=compaction2026.064
GLM-5.1 (attack_volume)Model=GLM-5.1, Condition=attack_volume2026.067
Claude-4.6 (compaction)Model=Claude-4.6, Condition=compaction2026.0619
Gemini-3.5f (attack_volume)Model=Gemini-3.5f, Condition=attack_volume2026.0619
GLM-5.1 (attack_inject)Model=GLM-5.1, Condition=attack_inject2026.0622
Qwen3.6 (attack_volume)Model=Qwen3.6, Condition=attack_volume2026.0626
Qwen3.6 (compaction)Model=Qwen3.6, Condition=compaction2026.0630
Qwen3.6 (attack_inject)Model=Qwen3.6, Condition=attack_inject2026.0633
Claude-4.6 (floor)Model=Claude-4.6, Condition=floor (no policy)2026.0633
DeepSeek-V4 (floor)Model=DeepSeek-V4, Condition=floor (no policy)2026.0637
GPT-5.4m (attack_volume)Model=GPT-5.4m, Condition=attack_volume2026.0637
GPT-5.4m (attack_inject)Model=GPT-5.4m, Condition=attack_inject2026.0637
Kimi-K2.5 (attack_inject)Model=Kimi-K2.5, Condition=attack_inject2026.0641
GPT-5.4m (compaction)Model=GPT-5.4m, Condition=compaction2026.0641
GLM-5.1 (floor)Model=GLM-5.1, Condition=floor (no policy)2026.0644
Kimi-K2.5 (attack_volume)Model=Kimi-K2.5, Condition=attack_volume2026.0644
DeepSeek-V4 (attack_volume)Model=DeepSeek-V4, Condition=attack_volume2026.0648
Qwen3.6 (floor)Model=Qwen3.6, Condition=floor (no policy)2026.0648
Kimi-K2.5 (floor)Model=Kimi-K2.5, Condition=floor (no policy)2026.0648
GPT-5.4m (floor)Model=GPT-5.4m, Condition=floor (no policy)2026.0656
DeepSeek-V4 (compaction)Model=DeepSeek-V4, Condition=compaction2026.0659
DeepSeek-V4 (attack_inject)Model=DeepSeek-V4, Condition=attack_inject2026.0659
Kimi-K2.5 (compaction)Model=Kimi-K2.5, Condition=compaction2026.0659
Gemini-3.5f (floor)Model=Gemini-3.5f, Condition=floor (no policy)2026.0663