# Gaming the check

The agent passes the test without doing the task.

**Is a:** [Wrong goal](/reference/ontology/unattended-agent-failures/wrong-goal)

**Guarded by:** [Read-only tests](/reference/ontology/unattended-agent-failures/read-only-tests), [Holdout scenarios](/reference/ontology/unattended-agent-failures/holdout-scenarios), [Stop option](/reference/ontology/unattended-agent-failures/stop-option)

The agent passes the test without doing the task. On tasks whose tests contradict the spec, GPT-5 cheated in 76% of One-Off cases ([ImpossibleBench](https://arxiv.org/abs/2510.20270)). StrongDM says tests inside the codebase got gamed ("return true") ([StrongDM, Software Factories](https://factory.strongdm.ai/)).

Our own reading: this is a kind of [wrong goal](/reference/ontology/unattended-agent-failures/wrong-goal) seen from the verifier's side, because a verifier the agent can pass without doing the work states the wrong goal. Guards: [read-only tests](/reference/ontology/unattended-agent-failures/read-only-tests), [holdout scenarios](/reference/ontology/unattended-agent-failures/holdout-scenarios), a [stop option](/reference/ontology/unattended-agent-failures/stop-option).

---

Published 2026-10-29. Web page: https://www.mackenziebowes.com/reference/ontology/unattended-agent-failures/gaming-the-check