RepoJournal
AI Red Teaming AI Red Teaming
24 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 2026-10-24
stories 34
RSS Archive Reviews

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

promptfoo 0.125.0 goes opt-in on Claude and Codex SDKs, PyRIT drops Gandalf

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

promptfoobot[bot] renovate hannahwestra25 richlundeen parkanzky promptfoobot romanlutz
10 people shipped this
promptfoobot[bot] @promptfoobot[bot] 1 cited renovate @renovate 1 cited hannahwestra25 @hannahwestra25 1 cited richlundeen @richlundeen 1 cited parkanzky @parkanzky 1 cited promptfoobot @promptfoobot romanlutz @romanlutz jbolor21 @jbolor21 Jeffrey Martin @jmartin-tech penguine @penguine-ip

Two breaking changes landed in the same window: promptfoo 0.125.0 makes the Claude Agent and Codex SDKs opt-in and demotes jks-js to an optional peer dependency, while PyRIT removes its Gandalf target and scorer.

0.125.0 promptfoo/promptfoo

by promptfoobot[bot]

Version 0.125.0 makes the Claude Agent and Codex SDKs opt-in and demotes jks-js to an optional peer dependency. Any eval config that relied on those providers resolving implicitly needs them declared before it runs; the release also adds provider-only environment overrides so credentials can be scoped per provider.

[BREAKING] MAINT Remove Gandalf target and scorer; make custom targets notebook self-contained microsoft/PyRIT

by hannahwestra25

The Gandalf target and scorer are gone from the main branch, so any attack or scoring config that referenced Gandalf will fail to load. The same change makes the custom targets notebook self-contained, so it runs without the removed pieces.

FEAT: Edit run-scoped technique settings inline microsoft/PyRIT

by richlundeen

Run-scoped technique settings are now editable inline, so tuning a technique for one run no longer means editing and re-saving the scenario definition. Changes stay scoped to that run rather than propagating to the stored config.

atomic plugin cache json file copy NVIDIA/garak

by parkanzky

Running multiple garaks from scratch at once used to race on the plugin cache copy and leave instances loading an incomplete JSON file; the copy is now atomic. The PR's reproducer hits it by repeatedly deleting ~/.cache/garak/resources/plugin_cache.json while instantiating PluginCache, which matters if you parallelize scans across containers.

chore(deps): update dependency oxc-parser to ^0.154.0 promptfoo/promptfoo

by renovate

Routine dependency work elsewhere: oxc-parser moves to ^0.154.0 and @hono/node-server to 2.1.5, with no operator-facing behavior described in either.

$ ls ai-red-teaming/reviews/ # the step-back reads

the week in review promptfoo 0.125.0 makes Claude and Codex SDKs opt-in
Full archive → Weekly reviews →

Quick answers

What shipped in AI Red Teaming on October 24, 2026?
Two breaking changes landed in the same window: promptfoo 0.125.0 makes the Claude Agent and Codex SDKs opt-in and demotes jks-js to an optional peer dependency, while PyRIT removes its Gandalf target and scorer. In total, 16 commits, 17 pull requests, and 1 releases landed.
Who contributed to AI Red Teaming on October 24, 2026?
10 developers shipped this update, including promptfoobot[bot], promptfoobot, renovate, hannahwestra25, romanlutz, richlundeen, jbolor21, and parkanzky, and 2 more.
What were the notable AI Red Teaming updates?
0.125.0, [BREAKING] MAINT Remove Gandalf target and scorer; make custom targets notebook self-contained, and FEAT: Edit run-scoped technique settings inline.
♥

$ log AI Red Teaming --recent

PyRIT closes a judge-spoofing hole, promptfoo accepts Langfuse v6 29 shipped PyRIT calms Rust false positives and fixes attacker-view reporting 48 shipped PyRIT fences judge transcripts against role forgery 57 shipped Garak fixes BEAST beam search direction, Promptfoo locks down OpenAI provider config 50 shipped

$ log --all # elsewhere on the wire

Alignment & Post-training GSM8K scoring parses answers as exact decimals Safety Evals Viewer plays terminal sessions, Kimi K3.1 Pro lands in 0.3.284

$ ls showcases/

Safety Evals Agent Security Interpretability Alignment & Post-training

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.