Repository navigation
We’re thrilled to announce the release of Infosys Responsible AI Toolkit v2.2.1. This update focuses on addressing critical security vulnerabilities, along with improvements in code coverage and stability, without introducing any major new features.
This release strengthens the overall security posture and reliability of the platform while ensuring a more robust and maintainable codebase.
Stay tuned as we continue to push the boundaries of Responsible AI innovation!
For information on module specific changes in this release please refer below:
responsible-ai-mfe:
Enhanced frontend with improved THOT/COV integrations, payload standardization, secure HTML rendering, and UI fixes. Updated README, licensing headers, and optimized API handling. Test coverage added to improve reliability.
responsible-ai-shell:
No functional changes in this release. Module remains stable with no updates. Test coverage improvements applied where applicable.
responsible-ai-backend:
No changes introduced in this cycle. Existing APIs and services remain unchanged. Test coverage improvements applied.
responsible-ai-admin:
Updated API configuration JSON to support new/modified endpoints. Ensures alignment with latest backend services. Test coverage added.
responsible-ai-Hallucination:
Updated requirements to maintain compatibility and stability. Minor dependency alignment performed. Test coverage added.
responsible-ai-explain:
No updates in this module. Existing explainability capabilities remain unchanged. Test coverage improvements included.
responsible-ai-llm-explain:
Updated documentation to enforce use of gpt-35-turbo-16k for Graph of Thought (GOT). Ensures correct deployment configuration. Test coverage added.
responsible-ai-img-explainability:
No changes in this release. Module continues to function as previously implemented. Test coverage added.
responsible-ai-llm:
README updated to mandate gpt-35-turbo-16k for GPT-3.5 deployments. Ensures configuration consistency across environments. Test coverage added.
responsible-ai-security:
This module has been retired. For reference or legacy implementation details, please refer to release 2.2.0 codebase.
responsible-ai-redteaming:
This module has been retired. For any prior functionality or implementation details, refer to release 2.2.0.
responsible-ai-model-detail:
No changes introduced. Module remains as-is with existing capabilities. Test coverage improvements added.
responsible-ai-reporting-tool:
No updates in this release. Reporting features remain unchanged. Test coverage added.
responsible-ai-moderationlayer:
Addressed Prisma and Sonar issues, upgraded dependencies, and improved model handling. Added test coverage and updated documentation/models for stability.
responsible-ai-moderationmodel:
Improved security by removing sensitive logs and addressing SAST findings. Added test coverage for robustness and validation.
responsible-ai-workbench/questionnaire:
Relaxed dependency pinning (pydantic, requests) to enable automatic secure updates. Improves maintainability and security posture. Test coverage added.
responsible-ai-privacy:
Strengthened security with cryptographic random generation, dependency upgrades, and vulnerability fixes. Updated README and improved compliance. Test coverage added.
responsible-ai-upload-doc:
Removed sensitive logging to mitigate system information exposure risks. Improved compliance with SAST standards. Test coverage added.
responsible-ai-safety:
Enhanced security by removing hardcoded keys and sensitive logs, and updating deprecated APIs. Added test coverage along with model updates.
responsible-ai-telemetry:
Implemented PII anonymization and added security headers for improved data protection. Fixed endpoints and expanded test coverage for reliability.
responsible-ai-fairness:
Major improvements including security hardening, dependency upgrades, bug fixes, and removal of deprecated services (BART). Added health endpoints and comprehensive test coverage.
responsible-ai-file-storage:
Improved security/config handling, fixed cloud initialization issues, and introduced environment-based configurations. Added extensive test suite and coverage.
responsible-ai-llm-benchmarking:
Enhanced security defaults, fixed critical bugs, and improved configuration handling. Introduced explainability benchmarking features and comprehensive test coverage.