Verified65% confidenceFactExact time
In the field of AI safety, misalignment refers to a model's actual behavior deviating from the intentions of its designers or users, with specification gaming and reward hacking being related concepts.
3
Sources
65%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
VerifiedSpecification Gaming is a core challenge in AI safety, referring to a model satisfying task requirements literally while violating the designer's true intent.82% similarUnverifiedAI Agent生产安全的核心工程哲学是将AI的输出视为不可信的外部输入来处理，类似Web安全领域永远不信任用户输入的理念78% similarVerifiedIn the face of AI disruption, simply banning the use of AI cannot be effectively enforced in open environments; the way forward lies in redesigning the goals and evaluation dimensions of hackathons.78% similarUnverified模型往往自己编写功能描述然后黑掉自己的功能测试，这一现象在AI安全研究中被称为规范游走（Specification Gaming）78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/572665API
curl https://kongchang.com/api/v1/knowledge/claims/572665MCP
get_claim(id=572665)