Skip to content

The guard test gets a real-looking address again, so main goes green - #3

Merged
leighanne77 merged 1 commit into
mainfrom
fix/guard-test-real-domain
Sep 28, 2026
Merged

leighanne77 merged 1 commit into
mainfrom
fix/guard-test-real-domain

Conversation

@leighanne77

Copy link
Copy Markdown
Owner

test_guard_blocks_real_email_in_tool_result has failed on every CI run since 2026-08-20. It proves the enterprise-mode guard refuses to send a prompt carrying an apparently-real email, but the de-identification pass moved its fixture to victim@megacorp.fake, and .fake is exactly what the guard is built to let through. The test was asking the guard to refuse a dummy address; the guard, correctly, did not.

The address is now victim@gmail.com. The de-identification gate already allows gmail.com for this reason in writing ("consumer-domain semantics under test"), two neighbouring tests use it, and the guard treats it as real, so the assertion means what it says again. example.com would not do: the guard counts it as dummy. No gate change.

test_guard_blocks_real_email_in_tool_result has failed on every CI run
since 2026-08-20. It proves the enterprise-mode guard refuses to send a
prompt carrying an apparently-real email, but the de-identification pass
moved its fixture to victim@megacorp.fake, and `.fake` is exactly what
the guard is built to let through. The test was asking the guard to
refuse a dummy address; the guard, correctly, did not.

The address is now victim@gmail.com. The de-identification gate already
allows gmail.com for this reason in writing ("consumer-domain semantics
under test"), two neighbouring tests use it, and the guard treats it as
real, so the assertion means what it says again. example.com would not
do: the guard counts it as dummy. No gate change.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@leighanne77
leighanne77 merged commit 4afc343 into main Sep 28, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant