[Bug] Content Safety Filter Incorrectly Downgrades Model for Legitimate Payment Security Analysis

**Bug Description**
你们的关键字审查与保护机制会导致我当前的项目被你们强制降级应用中的模型。例如：项目之前在fable5中进行分析后的出来的结论是：可能会有绕过付款，导致付款失败的问题。结果你们的关键字审查与防御的机制会强制要求降级模型，从而导致我的项目变成类似“非法”的项目，迟迟以最一般的模型推进。而低级别的模型会反复出现理解错误，让项目无限错误的走下去。让我的维护和推进的成本成几何倍的提升。已经遇到这个问题两周了。项目非但没有推进多少，还因为那个机制降级模型导致现在的项目需要不断的退回重做，退回重做的循环往复！

**Environment Info**
- Platform: win32
- Terminal: windows-terminal
- Version: 2.1.237
- Feedback ID: e8bd41a6-cadb-4e93-b89a-02c6913aae57

**Errors**
```json
[{"error":"TelemetrySafeError: VirtualMessageList: itemKeys/messages length desync (keys=88 messages=87 range=[0,88))\n at lGE (B:/~BUN/root/cli:23198:33898)\n at i8h (B:/~BUN/root/cli:23198:26870)\n at vi (B:/~BUN/root/cli:2140:21439)\n at Lm (B:/~BUN/root/cli:2140:40550)\n at Iy (B:/~BUN/root/cli:2140:51489)\n at ii (B:/~BUN/root/cli:2140:89194)\n at co (B:/~BUN/root/cli:2140:88149)\n at $t (B:/~BUN/root/cli:2140:87970)\n at kUe (B:/~BUN/root/cli:2140:84225)\n at gt (B:/~BUN/root/cli:2140:6694)","timestamp":"2026-09-01T15:50:26.127Z"}]
```

No source file, test, or entry point is named. Start by reproducing the model downgrade on win32 in Windows Terminal with version 2.1.237, using feedback ID e8bd41a6-cadb-4e93-b89a-02c6913aae57 and the reported VirtualMessageList desync. Done means legitimate payment-security analysis is not downgraded and the reported error no longer occurs.

索引モデルが issue の本文から書いたものです。