Cresec routine risk worksheet

List the AI routines your teams run in Claude, ChatGPT or Copilot. For each one,
mark the three legs. A routine with all three can be turned into a data leak by
one hidden instruction in the text it reads.

Routine | Owner | Assistant | 1. Reads outside text? | 2. Sees private data? | 3. Sends or changes something? | Legs
--------|-------|-----------|------------------------|-----------------------|--------------------------------|-----
        |       |           |                        |                       |                                |
        |       |           |                        |                       |                                |
        |       |           |                        |                       |                                |
        |       |           |                        |                       |                                |
        |       |           |                        |                       |                                |

1. Reads outside text: emails, tickets, CVs, call transcripts, web pages,
   documents anyone outside your company wrote.
2. Sees private data: CRM, billing, applicant records, contracts, tickets
   from other customers.
3. Sends or changes something: an email, a Slack message, a CRM field, a
   ticket, a file share.

Three legs: start here. Two legs: removing one leg, often by keeping it
read-only, may be enough. One or none: you probably don’t need Cresec for it.

Leave out credentials, customer data and confidential details.
https://cresec.ai/resources/