AI Transparency Notice

This public website

The public Trunnion AI marketing website does not provide an AI chatbot, generate personalized outputs for visitors, score visitors, or make employment, credit, housing, insurance, health, legal, government-benefit, procurement-award, security-clearance, or other consequential decisions. Optional analytics measures site use only after consent.

Product workflows vary by scope

Trunnion is presented as an orchestration and governance layer for configurable multi-agent workflows. The exact workflow, model and provider, prompt and tool inputs, retrieval sources, output recipients, human-review gate, logging, retention, training-use setting, deployment location, and availability must be confirmed in the current release documentation and written agreement. Marketing examples and interface illustrations are not proof that a feature is production-ready or enabled for a customer.

Limits and human oversight

AI outputs may be inaccurate, incomplete, biased, outdated, non-unique, insecure, or unsuitable. Human approval controls reduce risk only when they are correctly configured, assigned to qualified reviewers, usable under real operating conditions, and tested against bypass and failure. Customers remain responsible for defining prohibited uses, review thresholds, escalation, stop controls, complaint handling, correction, monitoring, and legally required impact assessments.

Consequential and regulated uses

No Trunnion product should be used as the sole basis for a consequential or regulated decision unless the exact use has been legally and operationally approved, appropriate human review is effective, relevant data and protected-class impacts have been evaluated, and required notices, appeals, records, assessments, and monitoring are in place. Website language is not a representation that a use complies with law or policy.

Data and model providers

Do not enter classified, CUI, export-controlled, confidential, personal, or regulated information unless the written scope and deployed controls expressly permit it. Whether a hosted or local model receives data, where processing occurs, whether data is retained or used for provider training, and which subprocessors have access must be established before production use.

Evaluation and provenance status

No portfolio-wide public accuracy, bias, robustness, prompt-injection, or gate-bypass evaluation result is currently claimed. Product claims must remain bounded until a dated evaluation method and result are available. Exported AI-assisted material is not represented as carrying a universal persistent technical marker; the responsible human reviewer remains accountable for the final artifact and should disclose material AI involvement where the audience could otherwise reasonably infer independent authorship or professional review.

Questions and reports

Questions about a specific workflow's provider, data use, limitations, human-review path, evaluation results, or incident handling may be sent to contact@viceroynm.com. Use the same route to report a harmful or unauthorized agent action, approval-gate bypass, suspected cross-tenant exposure, or materially incorrect consequential output. Contracted customers should use their documented support and incident channels.