> ## Documentation Index
> Fetch the complete documentation index at: https://docs.incident.io/llms.txt
> Use this file to discover all available pages before exploring further.
# AI governance
> The controls you can apply to AI usage, where each one lives, and what it affects.
Most AI controls live in [Settings → AI governance](https://app.incident.io/~/settings/ai-governance). The rest sit with the feature they control.
## Every AI control
| Control | Where to set it | Default | What it affects |
| ------------------------------- | -------------------------------------------------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------- |
| Store incident channel messages | **Settings → AI governance** | Enabled | Whether we keep incident channel messages. Most AI features need this on. |
| AI incident access | **Settings → AI governance** | All incidents and alerts | Whether AI features work in private incidents and alerts |
| AI data redaction | **Settings → AI governance**. Contact us to unlock it | Off | Strips credit card numbers, US Social Security numbers and phone numbers before content reaches a model |
| AI subprocessors | **Settings → AI governance**. Contact us to change | All allowed | Which providers may receive your data |
| @incident | **Settings → AI governance → Store incident channel messages** | Enabled on Pro and Enterprise | Chatting to the agent in Slack, the Microsoft Teams incident tab, the dashboard and the mobile app |
| Suggestions | **Settings → Suggestions** | Off | Suggested summaries, follow-ups and updates, and the catch-up posted to someone joining a channel |
| Investigations | **Investigations → Investigation settings** | Off | Whether investigations run, and [when](/investigations/triggering) |
| Proposing code fixes | **Investigations → Investigation settings** | Off | Whether an investigation opens a draft pull request without being asked |
| Code scanning, per repository | **Investigations → Code changes** | Off | Whether we read a repository's merge requests and use it as investigation context |
| Scribe | **Settings → Calls → Scribe** | Off | Whether [Scribe](/ai/scribe) joins incident calls, transcribes them and posts notes |
| Scribe auto-join | **Settings → Calls → Scribe** | On | Whether Scribe joins every incident call, or waits to be invited |
| Meeting Notes | **Settings → Calls → Scribe** | On | Whether people can add Meeting Notes to post-mortems to record debrief calls |
| Transcript viewing | Contact us | On | Whether people can read call transcripts in the dashboard and through the API |
| Auto-delete call notes | **Settings → Calls → Scribe** | Never | Deletes transcripts and notes 7, 14 or 90 days after the call |
| Remote MCP server | **Settings → MCP** | On | Whether AI clients such as Claude or ChatGPT can query your incidents, alerts and escalations |
## Who can change these settings
* The **Manage security settings** permission covers incident channel message storage, alongside your organization name and your SAML and SCIM configuration. By default, only account owners have this permission.
* The **Manage organization settings** permission covers everything else. Admin and Owner hold it by default.
You can grant either to a custom role. See [user roles and permissions](/admin/user-permissions).
To unlock AI data redaction or turn off transcript viewing, email [help@incident.io](mailto:help@incident.io).
Every change is recorded in your [audit log](/admin/audit-logs).
## Message storage
AI features use the incident channel conversation as key context, so this must be enabled to use most AI features. This includes the `@incident` agent in Slack, the web dashboard and mobile app, as well as Investigations.
Your data, including messages, is never used to train models.
## Private incidents and alerts
Even with private access allowed, a private incident never surfaces in search results or in suggestions for a different incident. The Slack agent is unaffected by this setting: it can be used anywhere in Slack, and never surfaces private incident or alert information.
## AI data redaction
[AI data redaction](/admin/managing-sensitive-data#ai-data-redaction) strips credit card numbers, US Social Security numbers and phone numbers from incident channel messages, the attachments AI reads and Scribe transcripts before they reach a model. Other content, including alert payloads, custom fields and code, is sent as written.
## Subprocessors
Your organization consents to OpenAI, Anthropic, Google Vertex, Recall.ai and ElevenLabs at sign-up, so all of them are allowed. [Settings → AI governance](https://app.incident.io/~/settings/ai-governance) lists their status once you've opted out of any. You can opt out of any of them: talk to your account manager, or email [help@incident.io](mailto:help@incident.io).
AI features route across whichever LLM providers you allow. Opting out of Google Vertex turns off code analysis. Opting out of ElevenLabs moves Scribe to your meeting provider's captions, and [voicemail](/on-call/live-call-routing#send-calls-to-voicemail) transcription to Twilio.
## FAQs
No. We hold zero data retention agreements with the LLM providers behind our AI features, OpenAI, Anthropic and
Google Vertex, so what we send them is not stored by them and is never used to train or fine-tune a model. The audio
providers behind Scribe and voicemail don't retain your data either. See [AI data handling](/admin/ai-usage).
Not currently. Our AI features use a combination of models across providers, chosen for accuracy, so there is no
bring-your-own-model or bring-your-own-key option.
Our [Trust Center](https://trust.incident.io/) holds the current subprocessor list, the data processing addendum,
and our SOC 2 Type II report.
## Related
Which providers we use, and how your data is handled.
Prevent, erase and redact sensitive information.