Breathe AI Assistant is an in-product AI chat experience that helps employees get answers to policy questions using company documents they already have permission to access. It is designed to return citation-backed responses and is intentionally scoped as a focused retrieval-based assistant rather than a general-purpose chatbot or autonomous agent.
This policy explains how Breathe AI Assistant is intended to be used, how we protect information, and what controls are in place to support safe and responsible use.
What does Breathe AI Assistant do?
Breathe AI Assistant helps employees ask HR and policy questions in natural language and receive answers based on company documents already available to them in Breathe. Responses are designed to include sources so users can refer back to the underlying document content.
How can I start using Breathe AI Assistant?
Breathe AI Assistant is being introduced under account-level feature control, with access limited to authenticated users inside their tenant. Availability can be managed without redeploying the product, which supports controlled rollout and rollback where needed.
How much does Breathe AI Assistant cost?
During Beta, rollout and usage are being actively managed through account-level enablement and usage caps on requests and tokens. This public policy does not set commercial terms, but it reflects that the service is being introduced in a controlled, monitored way while usage, cost and containment are evaluated.
How do I get help using Breathe AI Assistant?
Where a request is out of scope, Breathe AI Assistant is designed to direct the user back to the relevant Breathe experience or support path rather than attempting to answer beyond its intended boundaries. Product, Engineering, Support and Customer Success teams also review feedback and usage patterns to improve the experience over time.
Can my company opt out of Breathe AI Assistant?
Yes. Access to the AI Chat module is gated through an account-level feature flag, and the feature can be disabled for an account or user. Emergency rollback is also supported through global flag disablement.
Do I need technical expertise to use Breathe AI Assistant?
No. Breathe AI Assistant is intended to let employees ask normal HR and policy questions in everyday language and receive grounded answers from company documents, reducing the need for manual document searching or deep navigation through the product.
Where is my data stored and processed?
Breathe AI Assistant operates within Breathe backend services hosted in the existing AWS environment and uses Amazon Bedrock Knowledge Bases for retrieval-augmented generation. The knowledge base is backed by an Aurora PostgreSQL vector store managed in AWS, with Breathe controlling the metadata schema and retrieval configuration used to scope access.
How does Breathe protect my data?
Breathe AI Assistant reuses Breathe’s existing authentication and authorisation controls and applies deterministic retrieval scoping on account and user attributes for every query. Traffic between Breathe services and AWS Bedrock is protected with TLS, and the architecture is designed so the model does not receive content outside the authorised, filtered retrieval set.
How is staff personal and private information kept safe?
Breathe AI Assistant is designed to operate only on the documents and metadata necessary to answer a question, and in V1 it does not have direct access to core HRIS transactional data. It only sees documents the employee is already permitted to access, and personal documents or user-owned content are explicitly out of scope in V1.
Is there an audit trail for AI interactions?
Yes. Chat and retrieval events, including guardrail triggers and feedback, are logged to support auditability, monitoring and incident investigation. Breathe also maintains change history through versioned product and technical documentation as the system evolves.
How is Breathe approaching responsible AI compliance?
The Breathe AI Assistant assurance framework is intended to support ISO/IEC 42001 requirements covering governance, risk management, robustness, human oversight, transparency and documentation. Security, privacy and compliance controls are also aligned to the ISO 27001 and ISO 42001 Annex A control approach described in the assurance file.
How does Breathe reduce AI hallucinations?
Breathe AI Assistant uses a retrieval-augmented generation design so answers are grounded in retrieved company documents rather than free-form model memory. The service is designed to show sources, refuse unsupported or off-scope requests, and apply strict retrieval scoping so answers are based on authorised content only.
How is Breathe thinking about AI bias?
Breathe AI Assistant is designed as an informational service rather than an automated decision-maker, which materially reduces the risk of biased decision outcomes. It relies on document-based retrieval, existing visibility permissions, explicit non-goals around decision-making, and ongoing review of interactions to identify systematic issues or uneven outcomes.
What Breathe AI Assistant is not for
Breathe AI Assistant should not be relied on to make decisions about hiring, firing, performance ratings, disciplinary action or other people-related determinations. It is also not intended to provide legal or employment-law advice, process sensitive transactional HR data beyond authorised document access, or perform actions on a user’s behalf in V1.
Ongoing monitoring and improvement
Breathe monitors usage, cost, containment, guardrail behaviour and performance so that the service can be improved safely over time. Feedback, testing, feature flags and incident handling processes are used to support controlled change and rapid response where issues are identified.