Responsible AI
We believe that artificial intelligence should enhance teamwork, decision-making, and collaboration while keeping people, ethics, and trust.
Responsible AI
At TeamRetro, we believe that artificial intelligence should enhance teamwork, decision-making, and collaboration while keeping people, ethics, and trust at the heart of every interaction. Our commitment to responsible AI ensures that the tools we build are transparent, fair, and aligned with these values.
TeamRetro does not use your data to build or train AI models.
All content generated using TeamRetro’s AI tools is subject to our platform’s overall security, privacy, and terms of use.
TeamRetro AI capabilities
TeamRetro integrates generative AI into various features to enhance retrospectives, health checks, estimations, and standups. These AI features streamline your experience by answering questions through our AI Assistant (trAI), generating templates, health models, icebreakers, suggesting groupings, actions, providing summaries and insights, and tracking team sentiment and topics over time.
| Capability | Provider | Model |
|---|---|---|
| Retrospective template, icebreaker question, health model generation | Open AI | ChatGPT 5.x |
| AI Assistant (trAI) | AWS Bedrock | Claude Sonnet 5.x |
| Meeting summarization | AWS Bedrock | Claude Sonnet 5.x |
| Meeting suggested titles | AWS Bedrock | Claude Haiku 5.x |
| Insights topic, sentiment, keyword annotations. Theme identification. | AWS Bedrock | Claude Sonnet 5.x |
| Automated grouping, group title suggestions | AWS Bedrock | Claude Sonnet 5.x Claude Haiku 5.x |
| Action suggestions | AWS Bedrock | Claude Sonnet 5.x |
| Search embeddings | AWS Bedrock | Titan Text Embeddings v2 Cohere Embed Model 3 |
| AI image generation for cover and topic images | fal – Features & Labels, Inc. | Image-generation models hosted on fal.ai |
| Moderation of typed themes and image descriptions | Open AI | omni-moderation-latest |
Learn more about our AI-powered features. Learn more about our sub-processors.
Our principles
We are committed to responsible AI governance. Our AI use must:
- Be fair, transparent, and accountable.
- Avoid bias and discrimination.
- Use data ethically and lawfully.
- Be explainable to users, with clear documentation of decision-making processes.
- Comply with our SOC 2 Type 2, GDPR and other privacy rules.
Our commitment to safety
Principles are important, but so are practices. To make sure our AI stays aligned with your needs, we actively monitor, moderate, and restrict AI use where appropriate. These include, but are not limited to:
- The AI Assistant (trAI), summaries and suggestions run on AWS Bedrock under AWS Bedrock Guardrails. The guardrail blocks attempts to override our instructions (prompt attacks).
- The guardrail also screens AI output for violent, sexual, hateful, insulting or criminal content.
- The theme or image description you type for generated templates, health models, icebreaker questions and images is checked by OpenAI’s moderation model before anything is generated. Flagged requests are refused.
These safeguards cover TeamRetro’s own AI features. If you connect your own AI assistant, such as Claude, through our MCP server, it can read and change only what you can in TeamRetro. Your AI provider’s terms apply to the data it receives. Account administrators can turn off MCP connections for the whole account.
AI regulation and the EU AI Act
TeamRetro does not develop or train AI models, so the conformity assessments and conformance testing the EU AI Act requires of model providers do not apply to us. The models behind our features are built and operated by our providers: OpenAI, Anthropic and Amazon (via AWS Bedrock), Cohere, and fal.ai, who carry the model-level obligations, including those for general-purpose AI models.
Our own use of AI is assistive. Features like summaries, suggested groupings, and action suggestions support your team’s discussion, and a person reviews the output and decides what to keep. The AI Assistant (trAI) can also create and change items in a meeting, but only when a user asks it to, only within that user’s own permissions, and its changes stay visible and editable like any other. TeamRetro’s AI features do not make automated decisions about individuals, and based on our assessment they are not high-risk AI systems under the EU AI Act, so no conformity assessment is required for them either.
We monitor the EU AI Act and related regulation as obligations come into force, and we will update this page if our position changes.
Reporting and feedback
Responsible AI is not a one-time promise; it’s an ongoing journey. We continuously review, refine, and update our practices as technology evolves, ensuring our solutions remain aligned with the highest ethical standards.
If you encounter inappropriate, inaccurate, or unexpected AI content, or believe a prompt or output was incorrectly flagged, please contact our support team so we can investigate at info@teamretro.com.