CLAIM-2026-10-13-03

Active Deception is defined as models recognising infeasible or conflicting requirements but presenting them as achievable; Passive Deception is defined as models helping fulfil explicit user requests to mislead others.

Why it matters

This is the correction the edition turns on. The larger of the two rates is not the model deceiving its operator but the model complying with its operator's request to deceive a third party, which moves the exposure from provider to deployer.

Sources

  • arxiv2026swedecept