A negative test checks that a system correctly refuses something it should not allow. Instead of confirming that a permitted action works, it tries a forbidden one, such as reading another user's data or calling a blocked tool, and passes only if the attempt is rejected.
It matters for access control, because most tests only check that legitimate actions succeed. A permission system that allows too much can pass every ordinary test, and only negative tests reveal that a boundary is missing.
Writing them takes thought, since the number of possible forbidden actions is large and a test should be realistic. A negative test passing shows that the tried action was blocked, not that every similar action is. It needs to be run again when permissions change, and a test that fails for the wrong reason, such as a typing error, can give false confidence.
To use them, list the boundaries that matter, write a test for each, check that the refusal is for the intended reason and include the tests in automated runs.