LIMBO: Frontier Agents Duplicate Writes in 56-74% of Episodes When a Request Is Still in Flight or Delivered Twice

arXiv·high signalXBlueskyLinkedInCopy link

arXiv 2609.29095 ran 25,930 episodes across nine models, three production agent harnesses and twelve injected fault modes, and graded each episode against a ledger of committed side effects. When the agent is told to act exactly once and an immediate read-back can reveal what happened, frontier models almost never duplicate a lost-ack write (0.5%), and the model explains 53% of the variance. When the request is still in flight or the transport delivers it twice, the same models duplicate in 56% and 74% of episodes, and the tool contract explains 81%. The authors prove that no verify-only policy is exactly-once under late commits, so idempotency keys in the tool contract carry the guarantee, and prompting the model does not.