Unverified50% confidenceFactExact time
Safety alignment is not absolutely reliable, and models may fail due to out-of-distribution inputs, multilingual transfer, or prompt construction.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedUnlike traditional software security vulnerabilities, the boundaries of language model jailbreaks are inherently fuzzy, and the same output may be judged as safe or dangerous depending on the context.77% similarUnverifiedLarge language models currently have no reliable built-in defenses capable of identifying and rejecting prompt injection in all situations.75% similarUnverifiedDeterministic guardrails can only handle fixed-pattern content that looks like sensitive information and cannot detect natural language risks where sentences are fluent but subtly guide users toward incorrect actions.75% similarUnverifiedSafety-aligned language models have a one-dimensional direction in activation space associated with refusing harmful requests.73% similarUnverified确定性防护无法处理自然语言层面的隐晦风险，是其主要局限73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1103321API
curl https://kongchang.com/api/v1/knowledge/claims/1103321MCP
get_claim(id=1103321)