anthropics / anthropics/claude-code

[Bug] Anthropic API Error: Opus 5 Safeguards Incorrectly Flagging Legitimate Code Tasks

Đang mở
#91,257 0 bình luận 0 reaction 0 người được giao Xem trên GitHub
api:anthropic area:model area:security bug duplicate platform:windows
Ngôn ngữ chính
Python
Star
145k
Fork
23.1k
Chỉ số merge pull request
Chỉ số pull request đang chờ

Mô tả

I am currently pursuing a master’s degree in Artificial Intelligence at Dankook University, where my research focuses on cybersecurity and academic paper writing.

In this session, I am developing an industrial control system (ICS) simulator. Because the project involves implementing simulated cyberattacks in a controlled research environment, the use of cybersecurity terminology and descriptions of attack techniques is unavoidable.

However, Claude’s safeguards are repeatedly triggered when I attempt to implement these simulated attacks, which frequently interrupts or blocks legitimate academic work. I fully understand the importance of safety measures, but I hope the system can better distinguish authorized cybersecurity research conducted in a controlled simulator from requests involving malicious real-world activity. Improving this distinction would make Claude considerably more useful for legitimate cybersecurity research.

**Bug Description**
● Opus 5 (1M context)'s safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag legitimate coding, cybersecurity, and biology tasks. Switched to Opus 4.8. Send feedback with /feedback or learn more

Details: `[cyber]`
⎿ Tip: You can configure model switch behavior in /config

**Environment Info**
- Platform: win32
- Terminal: windows-terminal
- Version: 2.1.250
- Feedback ID: 290e6a8e-0eae-4aa2-bee5-ce64c0da13fe

**Errors**
```json
[{"error":"TelemetrySafeError: VirtualMessageList: itemKeys/messages length desync (keys=71 messages=70 range=[47,71))\n at jS (B:/~BUN/root/chunk-xpgs82q6.js:24:22792)\n at Pm (B:/~BUN/root/chunk-xpgs82q6.js:24:16104)\n at Ha (B:/~BUN/root/chunk-s9wqej6k.js:18:20034)\n at Du (B:/~BUN/root/chunk-s9wqej6k.js:18:38200)\n at Fh (B:/~BUN/root/chunk-s9wqej6k.js:18:84275)\n at yg (B:/~BUN/root/chunk-s9wqej6k.js:18:83275)\n at Ju (B:/~BUN/root/chunk-s9wqej6k.js:18:83104)\n at wh (B:/~BUN/root/chunk-s9wqej6k.js:18:79629)\n at Ve (B:/~BUN/root/chunk-s9wqej6k.js:18:6260)\n at Ye (B:/~BUN/root/chunk-s9wqej6k.js:18:4791)","timestamp":"2026-08-31T16:48:21.483Z"}]
```

Hướng dẫn đóng góp

Chưa lập chỉ mục được hướng dẫn đóng góp cho kho mã nguồn này

Hướng nghiên cứu

Start with the reported Opus 5 safeguard fallback and the VirtualMessageList error, using the provided Feedback ID and timestamp to reproduce the session. Inspect the generated chunk locations in the stack trace and the /config model-switch behavior. Done would require a clear, reproducible distinction between the safeguard interruption and the itemKeys/messages desync, with expected behavior agreed for legitimate research tasks.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.

Đánh giá

Công nghệ
bun
Lĩnh vực
cli, security
Loại issue
Lỗi
Độ khó
4/5
Thời gian dự kiến
3-5 ngày
Mức độ hoạt động
Sôi nổi
Độ rõ ràng
Cần làm rõ
Mức phù hợp với người mới
30/100

Nhận issue mới trong hộp thư của bạn

Bản tóm tắt ngắn những issue GitHub phù hợp với người mới.