AI Use Disclosure: Responsible AI
Last updated 15 September 2026.
SLATE VENTURES LTD t/a Volt26 (NZBN 9429053681034) 21F Gordon Rd, Wanaka 9305, New Zealand Contact: legal@volt26.ai
1. Overview
Volt26 is an AI-assisted platform built to help founders think, plan, and act more effectively. This disclosure explains how AI is used within the platform, which providers power it, what the AI can and cannot do, and what responsibilities you carry as a user.
We have written this in plain language. If you have technical questions, contact legal@volt26.ai.
2. AI provider and models
By default, all AI functionality in Volt26 is powered by Anthropic's Claude models, called directly via the Anthropic API. We do not route your data through any third-party AI broker, aggregator or middleware. If you choose to bring your own model key in Connections, that provider processes your prompts under your own agreement with it.
| Use case | Model |
|---|---|
| Standard agent interactions | Anthropic's Claude (Sonnet tier) |
| Higher-complexity reasoning (where available) | Anthropic's Claude (Opus tier) |
| Lightweight / fallback tasks | Anthropic's Claude (Haiku tier) |
Model versions may change as Anthropic releases updates. We will revise this disclosure when we change the default model tier.
3. Accuracy and limitations: outputs are not professional advice
AI-generated content can be inaccurate, incomplete, out of date, or confidently wrong. This is a known characteristic of large language models.
You must not rely on Volt26 outputs as:
- Legal advice
- Financial or investment advice
- Accounting, tax, or regulatory compliance advice
- Medical advice
- Any other advice for which a licensed professional is legally required
Outputs are provided for informational and planning purposes only. You are responsible for verifying any material fact, figure, or recommendation before acting on it. Where professional advice is needed, please engage a qualified professional.
4. Human oversight: you stay the decision-maker
Volt26 is built so that you remain in control of what your agents do.
As of the current platform version: Agents do work inside Volt26 on their own: they save research and artefacts, capture ideas, start research runs, draft campaigns and support replies, and queue proposals overnight. Anything outward (an email through your connected sending account, a social post, a campaign send) waits for your explicit Authorize, and every action is written to an audit ledger. Agents do not execute code, make purchases or write to your external systems.
As more outward capabilities are introduced (for example sending outreach through more channels), they are governed by an explicit autonomy model, not removed from your control:
- You set how much autonomy each agent has, by risk level, from advise-only, where every action waits for your approval, through to acting within limits you define.
- Actions that affect the outside world or are hard to reverse (sending, spending, publishing, deploying, or changing external systems) require your authorisation, unless you have explicitly pre-authorised that category of action.
- You can see what your agents have done and change their autonomy settings at any time.
Our commitment is not that agents will never act, it is that you remain the decision-maker: you set the controls, you stay informed, and meaningful or irreversible actions happen only with your authorisation. We will update this disclosure as these capabilities become available.
5. Your data and AI training
We do not use your inputs or outputs to train any AI model.
Specifically:
- Anthropic does not train its models on API customer inputs or outputs by default. This is governed by Anthropic's API usage policies.
- Anthropic's standard API retention period is approximately 30 days (used for safety and abuse monitoring). Content their safety systems flag can be kept for up to 2 years. We are actively applying for Zero Data Retention (ZDR) with Anthropic, which, if granted, would mean your inputs and outputs are not stored by Anthropic after the API call completes.
- We will update this disclosure and our Privacy Policy when ZDR status is confirmed.
For full details on Anthropic's data handling, see https://www.anthropic.com/legal/privacy.
6. Safety screening
We do not run a content filter of our own. Every agent call goes to Anthropic, whose usage-policy safeguards apply to the request and the response. What Volt26 adds is provenance: content that did not come from you (feeds, uploaded files, customer messages, web search results) is labelled as untrusted inside the prompt, and an action an agent proposes on a turn that carried such content is held for your Authorize rather than executed.
None of this makes AI outputs accurate or free from error. It is a safety and provenance layer, not a fact-checking layer.
7. Indirect prompt injection risk: how we bound it
When an AI agent processes content you have provided (for example, pasted documents, business descriptions, or research notes), that content could contain text crafted to manipulate the AI's behaviour. This is known as indirect prompt injection.
How we bound this risk:
- Actions stay behind your authorisation. Inside-app work an agent does on its own is reversible and logged. Anything outward runs through the autonomy and authorisation controls: anything that affects the outside world or is hard to reverse requires your approval, or a category you have explicitly pre-authorised. So even a successful injection cannot unilaterally cause a high-risk or irreversible action.
- Untrusted content is labelled. Anything that did not come from you is wrapped and named in the prompt, and an action proposed on a turn that carried it waits for your Authorize. We do not scan for injection phrases; that approach is easy to evade.
- We separate system instructions from user-provided content in our prompt architecture.
We do not claim this risk is eliminated. Avoid pasting content from untrusted third parties into agent prompts, review agent suggestions before acting on them, and report anomalous AI behaviour to legal@volt26.ai.
8. Your responsibilities as a user
By using Volt26, you agree that:
- You will review all AI outputs before acting on them, and exercise your own judgement.
- You will not attempt to misuse the AI, including attempts to generate harmful content, bypass safety controls, or extract confidential information about our systems or other users.
- You will not input personal information about third parties (employees, customers, individuals) into agent prompts beyond what is necessary for a legitimate business task, and only where you have a lawful basis to share that information.
- You accept that AI outputs may be wrong, and that Volt26 is not liable for decisions you make based on AI-generated content. See our Terms of Service for the full liability position.
- You will report unexpected, offensive, or concerning AI behaviour to legal@volt26.ai as soon as practicable.
9. Feedback and questions
We are committed to responsible AI development and welcome feedback on our AI practices.
Contact: legal@volt26.ai Privacy Officer: Samantha Rae, privacy@volt26.ai
Related documents: Privacy Policy · Acceptable Use Policy · Trust & Security Overview.