RepoJournal
AI Red Teaming AI Red Teaming
52 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 2026-11-16
stories 20

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

PyRIT adds remote OTLP trace fetching and tool-call ordering checks

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

PyRIT's scorer contract work reached operators this cycle: remote trace backends, order-sensitive tool-call assertions, and an opt-in prefill attack, all adoptable without patching the library.

FEAT: Fetch trace evidence from a remote OTLP/JSON backend microsoft/PyRIT

by WatchTree-19

The deferred remote adapter from the scorer contract proposal landed as OtlpJsonTraceClient, which fetches each trace by ID from an HTTP endpoint that returns OTLP/JSON, so you can point scoring at a live Tempo (or any OTLP/JSON) backend instead of loading traces into the local InMemoryTraceClient. Configuration is a url_template such as http://tempo:3200/api/traces/{trace_id} plus optional hea...

FEAT: Score the relative order of tool calls from trace evidence microsoft/PyRIT

by WatchTree-19

ToolsCalledInOrder is a new scorer condition (condition_type: "tools_called_in_order") that requires at least two distinct exact tool names and evaluates their relative order, not just presence. Because order lives in its own leaf, you compose it with the existing presence check in the scorer tree rather than rebuilding one scorer.

FEAT: Add prefill technique with native assistant continuation microsoft/PyRIT

by shashank03-dev

The opt-in prefill technique sits on PromptSendingAttack and sends the converted user objective followed by an exact assistant prefix, with scoring deciding success independently. Request support is enabled by default on Hugging Face, OpenAI Chat, OpenAI Responses, LiteLLM, and AzureML chat targets, while HTTP, Playwright, and Completion targets stay disabled unless you turn it on.

FEAT Add seed-prompt inspiration entry and contextual Chat return microsoft/PyRIT

by Rohitpro1

The seed-prompt entry point now lives beside the composer with an empty-Chat inspiration callout, and the originating Chat route survives catalog filters, example selection, paging, and return navigation. Direct explorer entry no longer preselects an unrelated saved destination, and the existing draft handoff covers text, conversation-editor turns, linked media, and setup.

MAINT: Bump the minor-and-patch group across 1 directory with 12 updates microsoft/PyRIT

by dependabot

The routine dependency train moved PyRIT's aiohttp to 3.14.5, fastapi to 0.162.2, pydantic to 2.14.3, and pyodbc to 5.5.0; promptfoo similarly bumped playwright to 1.65.0 and @biomejs/biome to 2.6.5, none of which change how you run either tool.

Quick answers

What shipped in AI Red Teaming on November 16, 2026?
PyRIT's scorer contract work reached operators this cycle: remote trace backends, order-sensitive tool-call assertions, and an opt-in prefill attack, all adoptable without patching the library. In total, 6 commits and 14 pull requests landed.
Who contributed to AI Red Teaming on November 16, 2026?
7 developers shipped this update, including renovate, mldangelo-oai, Michael, WatchTree-19, Rohitpro1, shashank03-dev, and dependabot.
What were the notable AI Red Teaming updates?
FEAT: Fetch trace evidence from a remote OTLP/JSON backend, FEAT: Score the relative order of tool calls from trace evidence, and FEAT: Add prefill technique with native assistant continuation.