Unverified50% confidenceFactExact time
密码学教授Matt Green发现重放未经修改的旧推理数据块（即使来自不同会话、账户或模型）也能通过验证
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Claude团队版发布，加密推理风波引发AI透明度争议
bilibiliAI-seeker6/24/2026
Related Claims
UnverifiedMatt Green inferred that both companies most likely use a single global key to encrypt all client reasoning data, which constitutes a key management anti-pattern.69% similarUnverifiedMatt Green推断OpenAI与Anthropic极可能都使用单一全局密钥加密所有客户端推理数据，属于密钥管理反模式67% similarUnverifiedAuthenticated encryption (such as AES-GCM mode) can detect whether ciphertext has been tampered with while ensuring confidentiality.59% similarUnverifiedThe call stack backtracking method locates the encryption critical point by setting a breakpoint at the request-sending location, progressively observing whether the encrypted x-tif value in the headers has been generated, and repeatedly tracing back.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/113652API
curl https://kongchang.com/api/v1/knowledge/claims/113652MCP
get_claim(id=113652)