RepoJournal
AI Red Teaming AI Red Teaming
16 wires and counting

$ follow AI Red Teaming

Keep up with AI Red Teaming in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-10-13
stories 81

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

PyRIT gets readiness gating and single-technique runs, promptfoo fixes repeat stability

By RepoJournal · Filed · About AI Red Teaming · Composed from the cited sources · methodology

PyRIT shipped operator-facing changes on three fronts (honest deployment readiness, a one-call technique API, and an attack analytics REST surface) while promptfoo corrected repeat-stability accounting for cached graders.

MAINT: Fail CoPyRIT deployment pipeline when initialization fails microsoft/PyRIT

by jsong468

CoPyRIT deployment previously could succeed even when PyRIT initialization failed, because /api/health only proves the web server is alive. A new uncached /api/ready endpoint returns HTTP 200 when the runtime can accept runs and sends and HTTP 503 otherwise, with runtime state and the serving ACA revision, and it is exempt from authentication, compatibility, and runtime-admission checks so a fa...

FEAT: Run one technique against an objective with technique.execute microsoft/PyRIT

by richlundeen

You can now run a single technique against an objective with technique.execute instead of wiring a full orchestrator flow. Useful for spot-checking one attack path in isolation during an assessment.

fix(eval): count cached graders in repeat stability promptfoo/promptfoo

by mldangelo-oai

Repeat-stability calculation now counts cached grading components in nested and mixed aggregates; affected groups omit confidence intervals but keep their final verdicts, target-cache flags, and accounting fields. Malformed historical components are ignored so imported evaluations still summarize and export.

FEAT: Attack analytics REST microsoft/PyRIT

by romanlutz

PyRIT added a REST surface for attack analytics, exposing campaign results to tooling outside the Python process. If your reporting pipeline currently scrapes local result files, this is the endpoint to wire in.

FIX: decode variation selector payloads next to emoji with presentation selectors microsoft/PyRIT

by bIackr0se

VariationSelectorSmugglerConverter(action="decode") returned '\x0f' instead of the hidden text whenever an ordinary emoji like ❤️ or ✔️ preceded the payload, because the decoder treated the emoji's U+FE0F presentation selector as the payload start and stopped at the next ordinary character. decode_visible_hidden had a related defect; both are fixed in this PR.

Quick answers

What shipped in AI Red Teaming on October 13, 2026?
PyRIT shipped operator-facing changes on three fronts (honest deployment readiness, a one-call technique API, and an attack analytics REST surface) while promptfoo corrected repeat-stability accounting for cached graders. In total, 36 commits and 45 pull requests landed.
Who contributed to AI Red Teaming on October 13, 2026?
5 developers shipped this update, including mldangelo-oai, bIackr0se, jsong468, richlundeen, and romanlutz.
What were the notable AI Red Teaming updates?
MAINT: Fail CoPyRIT deployment pipeline when initialization fails, FEAT: Run one technique against an objective with technique.execute, and fix(eval): count cached graders in repeat stability.