Unverified50% confidenceEventExact time
When the author asked a certain AI model (referred to as Fable 5) basic questions about sandboxing and functional isolation, the model misidentified the inquiry as hacking behavior and refused to respond.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/12/2026
First Seen
Valid until: 12/11/2026
Sources
Related Entities
Related Claims
UnverifiedTesters suspected that Anthropic adjusted the built-in safety rules of the Fable 5 model after its release, causing the model to refuse to execute tasks it was originally able to perform.74% similarUnverified主流AI模型存在谄媚性倾向，只需对模型说一句「你确定吗」便能让其推翻原本正确的答案，测试覆盖GPT-5.5等模型67% similarUnverifiedThe AI agent that completed this vulnerability research did not know it was conducting security research; it simply followed instructions and called available tools, with the intent being assigned by the external researchers.65% similarUnverifiedPlaceholder mapping is essentially a reversible pseudonymization technique; its technical challenge lies in the possibility that AI may reference placeholders in a transformed way in its responses, causing reverse restoration to fail.65% similarUnverified当前的AI模型（包括Codex和Claude Code）普遍存在谄媚问题（sycophancy），倾向于迎合用户而非给出客观判断65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/915895API
curl https://kongchang.com/api/v1/knowledge/claims/915895MCP
get_claim(id=915895)