Warrant uses artificial intelligence to review your marketing content against your configured policy library. This article explains what the AI does, how you and your team stay in control of every outcome, how we handle your data, and how our approach aligns with the EU AI Act and other governance frameworks.
You are interacting with an AI system
When you work in Warrant, the findings, severity ratings, and rationales labeled "AI" are generated by Warrant's AI system. We tell you this directly in the product the first time you use it, and every AI-generated finding is clearly and unambiguously marked so you can always distinguish AI output from content entered by people on your team.
What the AI does
When your team submits content for review, Warrant's AI compares it against the policies and regulations in your organization's policy library. For each potential issue it finds, it produces a structured finding made up of three parts:
The basis. The specific policy or regulation in your library that the finding relies on, cited so you can trace it back to its source.
A severity rating. An indication of how serious the potential issue is, to help your reviewers triage and prioritize.
A written rationale. A plain-language explanation of the reasoning behind the finding, so your reviewer can evaluate it rather than take it on faith.
Because each finding is grounded in a citation and an explanation, your reviewers can quickly judge whether it is well-founded, needs adjustment, or should be set aside. The AI's job is to surface and explain potential issues, never to reach a final verdict.
Humans make every decision
Warrant never auto-approves, auto-publishes, or auto-distributes your content. AI findings are recommendations only. A human reviewer on your team accepts, declines, or modifies every finding.
Every AI output passes through your team's review and adjudication before anything happens with it. As a result, content you publish after adjudication in Warrant reflects your organization's editorial control and compliance judgment, not an automated decision. This human-in-the-loop design is central to how Warrant works and is the reason the AI can be a genuine assistant without becoming a decision-maker.
Every finding is traceable
Warrant records an immutable audit trail of every AI finding. Each entry captures:
The policy or regulation cited as the basis for the finding;
The reference examples that informed it;
The severity rating and written rationale the AI produced; and
The action your reviewer took, accept, decline, or modify, and who took it.
These records cannot be altered by users, so you always have a complete history of how a decision was made.
How we handle your data
Warrant is designed so that your content and your reference material stay yours. A few principles govern how your data flows through the AI:
Isolation per customer. Each customer has its own isolated retrieval index. Your content never informs another customer's AI behavior, and no other customer's material informs yours.
No retention at the model layer. Inference calls to our LLM provider are not retained beyond the API call itself.
Admin-controlled reference material. Your admins control what reference material informs the AI in your tenant, and can request that it be pruned, versioned, or wiped at any time.
Your data is not used to train AI models
Warrant does not use your content to train its own models or any third-party AI models. Our LLM provider is contractually prohibited from training on customer data, and inference calls are not retained beyond the API call. Combined with the per-customer isolated retrieval index described above, this means your marketing content, your policy library, and your reviewers' decisions are never used to improve a model that another organization, or Warrant itself, would benefit from.
Known limitations
Warrant's AI is a review assistant, not a substitute for your compliance judgment. Being clear about what it cannot do is part of using it responsibly:
Like all AI systems, it can miss issues or flag items that your reviewers determine are not actually violations. That is exactly why human adjudication is mandatory rather than optional.
Findings are only as good as the library behind them. They are grounded in the policy library your organization configures, so the quality and coverage of your library directly shapes the quality of the findings. Keeping your policies current and well-structured is the single most effective way to improve results.
The AI does not know your business context beyond what is in the library and the content submitted. Reviewer judgment remains essential for nuance, intent, and edge cases.
Regulatory alignment
Warrant’s AI program is aligned with the NIST AI Risk Management Framework. Under the EU AI Act (Regulation (EU) 2024/1689), Warrant is classified as a Limited Risk AI system and meets the Article 50 transparency obligations: users are informed they are interacting with an AI system, and AI-generated content is clearly marked.
Details are available in our AI Policy — visit trust.hellowarrant.com or contact [email protected].
Questions?
Reach us at [email protected].