Data Use & AI Policy
Exactly how Maverik's agents use your data, what goes to AI model providers, and the controls you have.
Last updated 27 October 2026 · Applies to Maverik AI, a subsidiary and proprietary product of Aetheron Solutions.
1. Our principle
Maverik makes AI actions visible, controllable, reversible where possible, and tied to measurable outcomes. Your data is used to do your work, inside your workspace, and nothing else.
2. Data boundaries
- Workspace isolation: every record is scoped to a workspace and enforced at the database level with row-level security.
- Brand isolation: each brand has its own Brand Brain. Agents never retrieve memory from another brand or workspace.
- No cross-customer learning: patterns learned from your results improve only your workspace, unless you opt in to aggregated benchmarks.
3. AI model providers
- Agents send only the context a task needs, for example your offer and brand voice when writing ad copy.
- Customer personal data (names, phone numbers, emails of your leads) is kept out of prompts by default. Lead qualification can use it only when you enable it, and it is redacted where possible.
- We only use providers whose terms prohibit training on our customers' inputs and outputs.
- If we fall back to another provider during an outage, the same data boundary applies.
4. Training
We do not train or fine-tune models on your content, your brand data, or your customers' data without explicit, separate, revocable consent. Synthetic data is used for testing; development tools never receive customer personal data.
5. Auto mode and approvals
| Mode | What agents may do |
|---|---|
| Auto (default) | Run end to end: create, publish, manage ads and spend, and reply to leads within your caps and rules |
| Approval | Prepare actions; you approve before anything external happens |
| Suggest | Produce drafts only |
In every mode, spend caps apply per run, per day, and per workspace; brand and ad-policy checks run before anything goes live; failed actions are retried or rolled back; and a kill switch can stop any agent or the whole workspace.
6. Safety checks
- External content (websites, comments, competitor pages) is treated as untrusted and filtered for prompt injection.
- Each agent has an allowlist of tools it may use.
- Outputs go through policy and brand-compliance checks before they are published.
- Every run is traced: plan, tool calls, outputs, cost, and outcome.
7. Your controls
- Correct or delete any Brand Brain memory.
- Disconnect any platform; we stop using its tokens immediately.
- Export your workspace data.
- Delete a brand or workspace; data is removed within 30 days.