RepoJournal
AI Red Teaming AI Red Teaming
39 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 2026-11-08
stories 42

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

PyRIT fixes objective inference and adds set_value verdicts; promptfoo prunes Azure and Git dependencies

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

PyRIT's scorer stack gets two changes that touch how verdicts and objectives are computed, while promptfoo's day was almost entirely dependency-pruning refactors with no operator-visible behavior change.

FIX: Infer the objective from the user request when tool calls precede the answer microsoft/PyRIT

by Shizoqua

Any scorer that inferred its objective from a target's request got an empty string whenever the target ran a tool loop, because the extraction only read the turn at sequence - 1 and found the function_call_output there. Both the sync and async extractors now reuse the lookup rule from MemoryInterface.get_request_from_response, so multi-turn tool-calling targets report an objective instead of bl...

FEAT Set true/false verdicts through Score.set_value (phase 14) microsoft/PyRIT

by WatchTree-19

Score.set_value(value) now writes a true/false verdict and its status in one call, storing "true"/"false" and setting status alongside it, so a verdict can no longer drift out of step with its own status. Writing verdicts through score_value still works until 2.0, and float scores behave exactly as before, so nothing breaks on upgrade.

FIX: Remove TextTarget from GUI (#3777) microsoft/PyRIT

by alexcastgonz

TextTarget is gone from the GUI, so orchestrators that relied on the text target in the web interface no longer have that entry point.

MAINT Prefer public package imports and enforce export boundaries microsoft/PyRIT

by romanlutz

Imports in PyRIT now have to come through the public package paths, with export boundaries enforced. Custom scorer or orchestrator code that reached into internal modules may need its imports adjusted.

refactor(azure): replace the Foundry project SDK promptfoo/promptfoo

by mldangelo-oai

The promptfoo day was four dependency-swap refactors: Azure Foundry now uses the REST transport and existing OpenAI client, PFX signing calls the system OpenSSL with OPENSSL_BIN, Sheets reuses the existing Google auth library, and Code Scan uses native Git. Foundry, Sheets, Code Scan and signing behavior are documented as matching what they replaced; the only number that moved is the isolated r...

Quick answers

What shipped in AI Red Teaming on November 8, 2026?
PyRIT's scorer stack gets two changes that touch how verdicts and objectives are computed, while promptfoo's day was almost entirely dependency-pruning refactors with no operator-visible behavior change. In total, 19 commits and 23 pull requests landed.
Who contributed to AI Red Teaming on November 8, 2026?
5 developers shipped this update, including mldangelo-oai, alexcastgonz, Shizoqua, WatchTree-19, and romanlutz.
What were the notable AI Red Teaming updates?
FIX: Infer the objective from the user request when tool calls precede the answer, FEAT Set true/false verdicts through Score.set_value (phase 14), and FIX: Remove TextTarget from GUI (#3777).