This AI Addendum forms part of the Unifies Terms of Service at unifies.ai/terms and applies to the Service's artificial-intelligence features (“AI Features”), including transcription analysis, summaries, commitment and task extraction, search embeddings, and the Unifies assistant. Capitalized terms have the meanings in the Terms.
1. How the AI Features work
1.1. AI Features send content — including meeting transcripts, participant and workspace-member names and email addresses, and your assistant queries — to our third-party AI provider (currently OpenAI; see unifies.ai/subprocessors) over an encrypted connection, and return generated results. Transcription of meeting audio is performed by our capture provider (currently Recall.ai).
1.2. Unifies does not use Customer Content or Output to train its own models, and uses its AI provider solely to generate results for your workspace. Our AI provider processes Customer Content as our subprocessor.
1.3. Extracted commitments carry a confidence score and a source quote; a deterministic verifier checks quotes against the transcript and labels them exact, re-anchored, or unverified; low-confidence items are held for human review before delivery.
1.4. The assistant is read-only by default: actions it suggests (creating tasks, sending nudges, inviting members, exports) are executed only after a person confirms them, and the most consequential actions require typed confirmation.
1.5. Automatic task creation is configurable per workspace and can be disabled by an admin.
2. Customer obligations
2.1. Human review. Output can be inaccurate, incomplete, misattributed, or out of context. Customer must have a person review Output before relying on it or acting on it — including before treating an extracted commitment as agreed, assigning work, evaluating personnel, or communicating externally.
2.2. No high-risk use. Customer will not use AI Features: as the sole basis for decisions producing legal or similarly significant effects on a person (including employment, credit, housing, insurance, or medical decisions); for emotion recognition, biometric identification, or social scoring (this does not restrict use of the Service's built-in speaker-attribution features as provided); or in any manner prohibited by applicable AI, employment, or surveillance law. Execution records are evidence for human judgment, not a substitute for it.
2.3. Lawful inputs. Customer is responsible for its inputs, including the recording notices and consents described in the Terms (Section 4.2) and for not submitting prohibited data (Terms, Section 4.4).
3. Prompt injection and adversarial content
Meeting and message content is untrusted input. The Service fences and bounds untrusted content before model calls, caps and allow-lists everything written to connected tools, and requires human confirmation for assistant actions — but no mitigation eliminates prompt-injection risk. Customer should treat unexpected Output touching security, credentials, payments, or instructions to third parties with particular skepticism and report suspected manipulation to security@unifies.ai.
4. Output; rights; limitations
4.1. As between the parties, Customer owns Output (Terms, Section 6.2). Customer acknowledges that: similar or identical output may be generated for others from similar inputs; Output may embed third-party content from Customer's own inputs; and Unifies makes no representation that Output is original or non-infringing.
4.2. No professional advice. Output is not legal, financial, medical, tax, HR, or other professional advice.
4.3. Evidentiary limits. Receipts, transcripts, quotes, timestamps, and verification artifacts are working records produced by automated systems. Unifies makes no representation as to their admissibility, evidentiary weight, or sufficiency for any legal, regulatory, or employment purpose, and they are not certified records.
4.4. Accuracy disclaimer. AI FEATURES AND OUTPUT ARE PROVIDED “AS IS”. UNIFIES DOES NOT WARRANT THAT OUTPUT WILL BE ACCURATE, COMPLETE, CURRENT, OR FIT FOR ANY PARTICULAR PURPOSE. THE TERMS' WARRANTY DISCLAIMERS AND LIABILITY LIMITS APPLY FULLY TO AI FEATURES.
5. Changes
Unifies may change AI providers, models, and methods, subject to the subprocessor-change process in the DPA where Customer Personal Data is involved. Fair-use limits may apply to assistant usage to keep the Service reliable for everyone; Unifies will use reasonable efforts to notify Customer before enforcing a limit, and current allowances are available from legal@unifies.ai.