Unverified50% confidenceOpinionExact time
According to security researcher Jonathan Melton's analysis, the root cause of prompt injection lies in the design of session permissions, not in the model not being smart enough.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRelying on prompt constraints to prevent dangerous operations is inherently unreliable, because the model may be jailbroken, make misjudgments, or forget constraints during long conversations.70% similarUnverifiedThe core scenario that alignment researchers worry about is: sub-goals generated by an agent to achieve broad objectives may systematically deviate from true human intent, with errors cascading and spreading before being discovered.68% similarUnverifiedWhen building Scion, Anthropic made extensive use of adversarial review, that is, having the model actively look for flaws in the design.67% similarUnverifiedThe research argues that the shortcomings of large models in physical reasoning lie at the behavioral level (a lack of reliable iterative experimental decision-making capability) rather than at the knowledge level67% similarUnverifiedThe study points out a trade-off between safety and verisimilitude, where an overly broad definition of harmlessness makes models behave clumsily in adversarial scenarios.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063949API
curl https://kongchang.com/api/v1/knowledge/claims/1063949MCP
get_claim(id=1063949)