Exposure
Configuration exposure of each AI platform
Read-only, with the credentials you already gave Saible. Each finding comes with the script that fixes it. Built, but not yet run against a live account.
- OpenAIthe Admin key: projects, keys, model permissions and retention
- Anthropicthe Admin API key: workspaces, members and API keys
- Azure OpenAIthe service principal with the Reader role: accounts and deployments
- Amazon Bedrockthe connector's role with read grants: logging, guardrails and requests
- Snowflakethe reader role: Cortex settings, grants and Cortex usage
- Databricksa token or a service principal: serving endpoints and their usage
- Fix from hereA control on the platform itself. The finding includes the script, and a person runs it.
- Needs the gatewayOnly a control on the request path can address it.
- Platform consoleOnly the platform's own console or a commercial agreement can address it.
Steps
- 1Connect each platform's billing feed on Settings → Providers. Where a read needs an additional grant, the report names it.
- 2On Exposure, an organisation admin clicks Run now or waits for the nightly run.
- 3Open a finding to see the rows it is based on and the script that fixes it. Run the script yourself.
- 4An organisation admin clicks Acknowledge, or Resolve with a reason.
- 5Click Seal report. An organisation admin approves it in Review queue.
For your agents
This topic is also documented as Agent Skills that your agents can load: exposure-report. Request them from Saible.