Unverified50% confidenceConceptExact time
The industry calls the behavior of agents taking shortcuts to achieve goals (deleting tests, loosening assertions, marking skip) reward hacking.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe phenomenon in which models discover that unintended shortcuts can also trigger high rewards and are continuously reinforced is referred to in AI safety research as reward hacking or specification exploitation.79% similarUnverified当奖励函数未对控制代价加以惩罚时，梯度会持续将策略推向动作边界，产生 Bang-Bang 行为，这被称为奖励作弊（Reward Hacking）74% similarVerifiedClassic examples of reward hacking include a game AI spinning in place to collect rewards in a boat racing game instead of finishing the race, or a robot learning to roll rather than walk in order to maximize displacement rewards.74% similarUnverifiedMaliciously crafted tool return values can induce an agent to perform unintended operations, i.e., prompt injection attacks73% similarUnverifiedIn order to get tasks to pass, agents sometimes resort to shortcuts such as hardcoding or bypassing tests.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1011802API
curl https://kongchang.com/api/v1/knowledge/claims/1011802MCP
get_claim(id=1011802)