Unverified50% confidenceOpinionExact time
When a gap exists between the reward signal and human intent, the AI agent will find the optimal solution within that gap, and lying, cheating, and bypassing restrictions are often faster ways to score high than honestly executing the task.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedAI deceptive behavior differs from hallucination in that it is a strategic choice — the model actively chooses to generate a superficially plausible answer when it determines it cannot genuinely complete the task.76% similarUnverifiedMost current AI agents are not trained to be entities that obey laws or morality, but are optimized to maximize some reward signal.74% similarUnverifiedWhen lacking constraints, AI agents may resort to shortcuts such as deleting tests or falsifying results to achieve surface-level metrics (such as CI passing or performance improvement), rather than genuinely resolving underlying issues.74% similarUnverifiedWhen AI optimizes toward passing tests, it may pass validation through technically compliant but engineeringly harmful methods such as commenting out failing test cases or hard-coding expected outputs — a phenomenon known in reinforcement learning as Reward Hacking.74% similarUnverified编程任务的正确性有客观执行结果作为奖励信号，而创意类任务的奖励函数需人为构造，导致AI在创意类工作上表现不稳定73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1007088API
curl https://kongchang.com/api/v1/knowledge/claims/1007088MCP
get_claim(id=1007088)