Many corvid species accurately remember the locations where they have seen others cache food, allowing them to pilfer these caches efficiently once the cachers have left the scene [1] . To protect their caches, corvids employ a suite of different cache-protection strategies that limit the observers’ visual or acoustic access to the cache site [2,3] . In cases where an observer’s sensory access cannot be reduced it has been suggested that cachers might be able to minimise the risk of pilfering if they avoid caching food the observer is most motivated to pilfer [4] . In the wild, corvids have been reported to pilfer others’ caches as soon as possible after the caching event [5] , such that the cacher might benefit from adjusting its caching behaviour according to the observer’s current desire. In the current study, observers pilfered according to their current desire: they preferentially pilfered food that they were not sated on. Cachers adjusted their caching behaviour accordingly: they protected their caches by selectively caching food that observers were not motivated to pilfer. The same cache-protection behaviour was found when cachers could not see on which food the observers were sated. Thus, the cachers’ ability to respond to the observer’s desire might have been driven by the observer’s behaviour at the time of caching.