衡量前沿 AI 智能体在真实任务中遵守用户边界的可靠性,覆盖未授权操作、虚假归因、欺骗性完成与典型越界模式。
| Rank | Model / Lab | Alignment Index | Unauthorized action | False attribution | Deceptive completion | Samples |
|---|
How models fail
The most common failure modes for each model and signal. Percentages describe the mix of each model's flagged cases, and a case can belong to more than one mode.
| Model | Unauthorized cleanup | Unrequested output | Premature execution | Scope expansion | Protected-content rewriting |
|---|---|---|---|---|---|
| GPT-6.1 Sol | 14.3% | 14.3% | 14.3% | 28.6% | 7.1% |
| GPT-6 Astra | 16.0% | 20.0% | 16.0% | 24.0% | 8.0% |
| GPT-6 Luna | 35.3% | 23.5% | 23.5% | 11.8% | 5.9% |
| GPT-6 Sol | 21.4% | 25.0% | 21.4% | 10.7% | 7.1% |
| GPT-5.6 Sol | 45.2% | 14.3% | 23.8% | 19.0% | 9.5% |
| Anthropic Claude Opus 5.5 | 20.0% | 40.0% | 10.0% | 0.0% | 20.0% |
| Grok 4.7 | 30.0% | 23.3% | 30.0% | 10.0% | 6.7% |
| Grok 4.6 | 45.3% | 20.3% | 18.8% | 14.1% | 7.8% |
| Anthropic Claude Fable 5.1 | 6.5% | 16.1% | 38.7% | 16.1% | 12.9% |
| Anthropic Claude Sonnet 5.5 High | 32.0% | 14.0% | 18.0% | 26.0% | 8.0% |
| Gemini 4 Argon | 25.0% | 20.8% | 33.3% | 25.0% | 16.7% |
| Tencent HY4 Preview | 37.5% | 40.0% | 22.5% | 20.0% | 2.5% |
| Meta Muse Wonder | 30.0% | 28.3% | 21.7% | 20.0% | 15.0% |
| Meta Muse Spark 1.3 | 22.7% | 13.6% | 38.6% | 22.7% | 2.3% |
| Gemini 3.8 Flash | 27.7% | 29.2% | 26.2% | 23.1% | 18.5% |
| Anthropic Claude Sonnet 5 | 38.0% | 20.0% | 26.0% | 12.0% | 12.0% |
| Anthropic Claude Opus 5 | 53.5% | 32.4% | 28.2% | 22.5% | 4.2% |
| Reflection AI Beam | 25.0% | 25.0% | 35.0% | 30.0% | 20.0% |
| Stepfun Step-5 Preview | 29.8% | 31.9% | 23.4% | 21.3% | 2.1% |
| GLM 5.3 | 26.0% | 27.4% | 19.2% | 20.5% | 4.1% |
| Kimi K3 | 33.8% | 22.1% | 13.2% | 22.1% | 5.9% |
| DeepSeek V4.1 Flash | 32.8% | 39.1% | 26.6% | 18.8% | 7.8% |
| MiMo V2.6 Pro | 25.0% | 7.5% | 30.0% | 22.5% | 10.0% |
| Anthropic Claude Opus 4.8 | 30.4% | 26.1% | 18.8% | 15.9% | 14.5% |
| Gemini 3.7 Flash | 13.6% | 21.6% | 29.6% | 29.6% | 20.0% |
| Gemini 3.6 Flash | 13.1% | 33.3% | 31.3% | 24.2% | 20.2% |
| Qwen3.8 Max | 31.2% | 21.2% | 22.5% | 22.5% | 10.0% |
| Qwen3.8 27B | 24.6% | 24.6% | 30.8% | 21.5% | 7.7% |
| Thinking Machines Inkling | 22.0% | 30.0% | 22.0% | 29.0% | 23.0% |
| MiniMax M3 | 26.9% | 26.0% | 24.0% | 27.9% | 25.0% |
Unauthorized cleanup
The agent deletes the user’s files or earlier work without permission.
Example: Deletes the user’s original images after editing copies of them.
Unrequested output
The agent produces something the user did not ask for, or keeps working after being told to stop.
Example: Generates an image when the user asked only for the prompt text.
Premature execution
The agent starts making changes before the user has approved them.
Example: Edits a file when asked to assess a proposed change.
Scope expansion
The agent changes parts of the work the user did not ask it to touch.
Example: Rewrites the body when asked only to change the title.
Protected-content rewriting
The agent changes content the user explicitly said to keep.
Example: Alters facts during a grammar-only edit.