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 2026-10-02
stories 65

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

Promptfoo drops Slack from default installs

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

Promptfoo's Slack SDK is now opt-in, a breaking change for anyone whose config or CI installs the package and expects Slack scanning to work out of the box, and PyRIT's LiteLLM target gains opt-in trace propagation.

perf(deps)!: make Slack SDK opt-in promptfoo/promptfoo

by mldangelo-oai

Ordinary installs no longer pull in the Slack SDK or the unused direct ANSI tokenizer dependency, so a Slack-based scan that worked yesterday will now fail until you co-install `@slack/web-api@^8.1.1`; selecting Slack validates the installed SDK and prints installation guidance for missing or incompatible versions. The provider docs and shipped Slack example now use co-installation commands and...

FEAT: opt-in trace propagation for LiteLLMChatTarget (part of phase 6.5) microsoft/PyRIT

by WatchTree-19

LiteLLMChatTarget can now propagate traces, opt-in, as part of the phase 6.5 work. Existing runs keep their current behavior unless you turn it on.

fix(providers): honor per-call cache overrides promptfoo/promptfoo

by he-yufeng

AI21, Cohere, and LocalAI chat and completion providers now honor per-call cache overrides: `bustCache` takes precedence over the legacy `debug` fallback, including an explicit `bustCache: false`, while default caching is unchanged. If you were relying on `debug` to force a fresh call, check which flag your config actually sets.

fix(storage): retry Windows blob publication contention promptfoo/promptfoo

by mldangelo-oai

Concurrent local media stores on Windows were failing intermittently when an atomic rename hit a briefly locked destination. Data and metadata publication now retries `EPERM`, `EACCES`, and `EBUSY` with five bounded waits totaling 750 ms per file.

fix(cli): silence internal decompression warning promptfoo/promptfoo

by mldangelo-oai

The CLI no longer prints `ExperimentalWarning: DecompressInterceptor is experimental and subject to change` when its startup update check creates an HTTP agent; the notice is suppressed only while undici creates its decompression interceptor, and decompression behavior, header handling, and the decompressed-size limit are unchanged.

MAINT Python CI performance microsoft/PyRIT

by romanlutz

Elsewhere: PyRIT reworked its Python CI, garak corrected a misleading comment in a regression test after confirming that `Probe._execute_attempt` deep-copies attempts so sibling detector scores were never overwritten, and a breaking upstream commit landed in PyRIT.

Quick answers

What shipped in AI Red Teaming on October 2, 2026?
Promptfoo's Slack SDK is now opt-in, a breaking change for anyone whose config or CI installs the package and expects Slack scanning to work out of the box, and PyRIT's LiteLLM target gains opt-in trace propagation. In total, 31 commits and 34 pull requests landed.
Who contributed to AI Red Teaming on October 2, 2026?
10 developers shipped this update, including mldangelo-oai, he-yufeng, renovate, Richard Lundeen, WatchTree-19, romanlutz, behnam-o, and Pujitha Paladugu, and 2 more.
What were the notable AI Red Teaming updates?
perf(deps)!: make Slack SDK opt-in, FEAT: opt-in trace propagation for LiteLLMChatTarget (part of phase 6.5), and fix(providers): honor per-call cache overrides.