Evidence / Practical AI Safety

Sources and scope

The examples are fictional teaching cases. They are not transcripts, customer results, or security tests of a live assistant.

Sources reviewed September 21, 2026. Product controls can change. Follow the current provider guidance and your workplace rules.

Does ChatGPT tell the truth?

ChatGPT data controls

Memory in ChatGPT

Temporary Chat

OWASP: excessive agency

OWASP: AI agent security

How to use this material

The original guide and worksheets are available under the MIT license in the repository. Linked sources retain their own terms. This guide is independent of OpenAI and OWASP and does not imply their endorsement.

Corrections

Use the repository issue form to report a public factual error. Do not include private chats, access keys, or client information.