Skip to content

v1.1.0

Latest

Choose a tag to compare

@varunj-msft varunj-msft released this 04 Sep 22:19

What's changed?

PyRIT v1.1.0 improves scoring, expands scanner capabilities, and adds major CoPyRIT features. Python support remains >=3.10, <3.15.

⚠️ Breaking changes

  • Scores can now be undetermined. Check score.is_undetermined before reading score_value or calling get_value() (#2491).
  • Partial blocked responses are now scored by default. Set should_score_blocked_content=False to retain the previous behavior.
  • Authenticated CoPyRIT deployments now use Microsoft Graph and require an allowed group configuration (#2281).
  • The new fixed-egress CoPyRIT deployment requires a parallel Azure Container Apps environment. Azure SQL deployments also require the new migrations (#2438).

🧩 Framework, identity, and memory

  • Added explicit score status and scorable identity, with the related memory migrations.
  • Added WebSocket targets, round-robin adversarial targets, target reset hooks, and stop-reason metadata.
  • Improved CLI scenario results, remote authentication, Key Vault integration, partial-response handling, and attack-result output.

🎯 Scanner

  • Added Best-of-N, split-payload, multilingual, audio achilles, package hallucination, system-prompt extraction, and FigStep capabilities.
  • Added 10 converters, including bijection, CharNoise, CodeAttack, SATA, Vigenere, Acrostic, and IPA converters.
  • Improved package hallucination coverage, TAP lineage, converter isolation, and asynchronous execution.

📊 Scoring

  • Added support for undetermined scores and scoring partial blocked responses.
  • Added Roblox PII, ShieldGemma, and system-prompt extraction scorers.
  • Fixed aggregation, score metadata preservation by category, non-finite value handling, package detection, and plagiarism matching (#2579).

📚 Datasets

  • Added 488 TrustAIRLab in-the-wild jailbreak templates and the Menz et al. jailbreak datasets.
  • Added dataset refresh support and safer remote dataset caching.

🖥️ GUI

  • Added a scenario catalog and support for configuring and launching scenarios directly from the GUI (#2373).
  • Added configuration editing, GUI initializers, converter configuration, and HTML conversation export. Default deployments now initialize both target and attack technique registries (#2580).
  • Improved score loading, Markdown display, large-history performance, accessibility, and mobile support.

🔒 Security

No known high-priority security issues.

  • Moved CoPyRIT authentication and group authorization to Microsoft Graph.
  • Constrained HTTPXAPITarget upload paths and made dataset cache updates atomic.
  • Updated dependencies with available security patches.

Full list of changes

New Contributors

Full Changelog: v1.0.1...v1.1.0