Skip to content

Black Friday:

60% off

Cliedesk Pro

Every plan, every site count. Ends Cyber Monday.

00 days 00 hrs 00 min 00 sec
Claim 60% off

AI

AI governance

This guide covers a feature of Cliedesk Pro. The free plugin is a complete help desk; see what Pro adds.

AI governance is part of Cliedesk Pro. People with the Manage AI permission (Admins) open Settings → AI → Governance. It has four tabs: Processing notice, Register, Disclosure and Safety.

Accept the AI processing notice

Before Cliedesk sends anything to an outside AI provider, an admin must accept the AI processing notice. Until then, no external AI call happens (also no key test). AI on your own server needs no notice.

  1. Open Settings → AI → Governance → Processing notice.
  2. Read the list: each provider, how long it keeps data, and the data each AI feature sends.
  3. Tick I accept the AI processing notice for these providers and select Accept.

The acceptance is kept in the audit log. It covers the providers it lists. When you connect a new provider later, it shows New since acceptance, and Cliedesk sends nothing to it until you accept again (here, or with the box on Settings → AI → Providers when you connect it). Copy the Text for your privacy policy into your privacy policy. It is also in the WordPress privacy policy guide.

The compliance register

Register lists each AI provider and model, the region, the tasks and features that use it, and the data sent after masking. It updates itself when you change AI settings. For each provider, keep the status of your data processing agreement (DPA): Signed, Pending, Not required or Missing, with the date and a reference, then select Save. Select Download Markdown or Download PDF for a DPIA or a customer questionnaire. The register is generated from your settings. It is not legal advice.

Tell customers when an AI writes

The AI agent tells customers that they talk to an AI. It is on by default: the AI label on the first chat reply, the footer of AI emails and the spoken notice on calls. You change the wording per brand in Settings → AI → AI agent (see AI agent). On Disclosure you can change the AI label. When the AI voice agent is set up, you can also change the Spoken disclosure on calls: what the AI says first on a call. It must say that an AI speaks and that the call is transcribed. Emptying it needs the same acknowledgement and reason as removing the disclosure.

To remove the disclosure, switch off Tell customers when an AI writes to them, tick I acknowledge the compliance warning, write a Reason and select Save. Laws such as the EU AI Act require this notice. Your acknowledgement and reason are kept in the audit log. Every other change of the disclosure is logged too. The general settings screens cannot remove it. On Settings → AI → AI agent → Voice you can switch it off too: you confirm your password, confirm the warning and write a reason, and it is logged the same way. For calls, the record keeps the words the AI voice agent really says first.

Under Record of a customer, enter a contact ID and select Show to see which AI notice the customer saw, on which channel, whether they went on with the AI and whether they asked for a person. Select Download CSV to export it. The record is part of the WordPress personal data export and is deleted when the contact is erased.

The safety layer

The AI safety layer is always on. You cannot switch it off.

  • Customer text is treated as data, never as instructions. Attempts to change the AI’s instructions are logged.
  • Spam and phishing never reach the AI agent or other AI features that answer customers.
  • Answers are checked before anyone sees them: no raw HTML, no images from other sites, no dangerous links, no full card numbers. Email addresses and phone numbers the AI was not given are noted. Public answers that contain them are not posted, and the AI agent does not send such a reply in chat, email or any other channel: a person takes over the conversation. Links in the data the AI sends to an MCP server are noted when the AI was not given them. A tool the AI was not offered never runs.
  • If the safety layer cannot run, AI answers stop until it runs again.

Safety shows what was caught in the last 7 days. More than 10 injection attempts in an hour show a Spike note. Supervisors with the View AI traces permission see this tab’s events.

AI spam and phishing scoring

Cliedesk scores every customer message for spam. With Pro, the AI checks messages that look suspicious and the score uses both. A new conversation above the spam limit goes to Spam. Nothing is deleted. When you mark a conversation Spam or Not spam, the filter learns from it (after 3 of each). You can turn the AI check off with the AI spam and phishing scoring switch in Settings → AI → Switchboard.