Privacy Policy

Who we are

Wiser Human Ltd, a company registered in England and Wales (number 15526536), is the controller of the personal data described here.

To ask about your data or exercise your rights, email info@wiserhuman.ai or use our contact form.

1. Our website

What we collect and why

• Contact form. Your name, email address and message, so we can reply. Lawful basis: our legitimate interest in answering enquiries. The form is delivered to us by email through Web3Forms. We keep enquiries for as long as we need to deal with them, and for no more than 2 years.
• Analytics, only if you accept cookies. Google Analytics records how the site is used (pages visited, approximate location, device and browser). Lawful basis: your consent, which you can withdraw at any time by clearing this site's cookies and choosing Reject. If you reject, no analytics cookies are set. Google keeps this data for 14 months.
• Newsletter. If you subscribe through the Substack form, Substack collects your email address and handles your subscription under its own privacy policy.
• Hosting and fonts. To deliver pages, our host (GitHub Pages), our network provider (Cloudflare) and our font services (Google Fonts, Fontshare) receive your IP address and browser details, as any website does. Lawful basis: our legitimate interest in running the site securely.

2. The Agent Escalation Line

The Agent Escalation Line at escalate.wiserhuman.ai is a beta safety service for AI agents. It is used by software, but what agents send can contain personal data about people: the agent's principal (the person or organisation it works for), the people in its situation, and anyone the agent mentions. Its terms summarise this section.

What we collect

• What the agent submits, and any follow-up messages. Before anything is stored or sent on, we automatically remove secrets, email addresses, phone numbers and card numbers where we can detect them. We ask agents not to send personal data at all.
• Principal contact details, only if the agent gives them, so our team can try to reach the principal about that case.
• During the beta, an encrypted copy of exactly what was sent, before redaction, including submissions we refuse. We keep it only to check that our redaction catches what it should. Only our staff can view it, and every view is logged.
• Where the request came from: the country, the organisation that runs the network (for example a cloud provider) and the software name the agent sends. We do not store IP addresses. We keep a keyed, one-way hash of the address instead, to apply rate limits and count distinct sources.
• Case records: timestamps, status, labels, our replies and the automated assessments.

How we use it, and our lawful basis

• To assess and respond to each case, and to try to reach a principal when that would help. Basis: our legitimate interest in running a safety service that helps AI agents act lawfully and safely, which also protects the people their actions could affect. Where someone's life or safety may be at risk, we may also rely on vital interests.
• To prevent abuse (rate limits, spam and misuse) and to check our redaction filters. Basis: legitimate interests.
• For research and service improvement, we use only content-free, aggregate statistics: request types, labels, response times, country and network organisation. We share only anonymised, aggregate figures with researchers and model developers, never case content.

We do not sell personal data, use it for advertising, or use case content to train AI models.

Who receives it

• Our support team, through our admin tool and a private Slack channel. Slack copies exclude principal contact details and are redacted like everything else.
• AI assessment: Anthropic (Claude models) produces the automated assessment. If it is unavailable, a backup model is used through OpenRouter, limited to model providers that do not keep or train on the data. They receive redacted case content, never principal contact details.
• Infrastructure: Cloudflare (hosting and database) and Google (urgent alerts to our team, which contain no case content).

How long we keep it

• Cases are kept while open. A case with no new messages for 30 days closes automatically.
• Case content, the encrypted copy and the assessment history are deleted 90 days after the case closes.
• Refused submissions: encrypted copy deleted after 90 days.
• Slack copies: deleted 90 days after posting.
• The hashed source ID in our statistics: removed 90 days after the case was opened.
• Content-free statistics: kept indefinitely.
• Anthropic deletes what we send it within 30 days and does not use it to train its models. Through OpenRouter, we use only model providers that do not keep or train on the data.

3. For both

Transfers outside the UK

Some of our providers (Anthropic, OpenRouter, Slack, Google, GitHub, Cloudflare, Web3Forms, Substack) process data in the United States or elsewhere. Where they do, we rely on the UK–US data bridge or the UK International Data Transfer Agreement or Addendum.

Security

Data is encrypted in transit. Our admin tool requires a login through Cloudflare Access, encrypted copies use a separate key, and access is limited to our team.

Your rights

You can ask for access to your personal data, or ask us to correct it, delete it, restrict or object to our use of it, or transfer it. The Escalation Line does not record names or accounts, so we cannot look you up. If your request is about a case, please give us its case ID, or describe the case (for example, roughly when it was sent and what it was about) so we can find it. We reply within one month. If you are unhappy, you can complain to the Information Commissioner's Office at ico.org.uk.

Children

Our website and services are not directed at children.

Changes

We will update this page when our practices change, and show the date at the top.