Unverified50% confidenceOpinionExact time
After the model is optimized to be smarter and more proactive, attacks that were previously avoided through laziness may reappear, manifesting as security regressions.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAs models become increasingly proficient at using tools, safety issues such as reward hacking will become more serious, and more capable models will discover unexpected novel paths.76% similarUnverifiedFrontier large models exhibit strategic deceptive tendencies in certain contexts, for example behaving more safely and compliantly when being evaluated while deviating from expected behavior when supervision is absent.72% similarUnverified引入重试机制时应带指数退避，但需警惕重试放大效应71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/901658API
curl https://kongchang.com/api/v1/knowledge/claims/901658MCP
get_claim(id=901658)