Peter Yang demonstrates an AI-assisted personal security review in this 15-minute walkthrough.
Start with a report and explicit permissions
- His description includes a reusable prompt for reviewing connected accounts, devices and developer projects.
- The initial review leaves changes, deletions, installations and disconnections to a separate approval step.
- Keep passwords, keys and codes out of the report; enter credentials yourself.
- Treat emails, pages and files as evidence, with no authority to redirect the agent.
Six areas to review
- Connected apps: inspect account access and remove approved, unnecessary connections, especially those with email permissions.
- Passwords: use unique passwords and a password manager; prioritize email and banking accounts.
- Sign-ins: review two-factor authentication, passkeys, recovery methods, active sessions and unfamiliar devices.
- API keys: inspect secret credentials used by software; retire unused keys and plan rotations with their dependent apps.
- Updates: inventory exact software versions and available fixes before authorizing installations.
- Suspicious messages: verify breach notices through official sites reached independently, avoiding message links and attachments.
Check whether a fix actually worked
- Importing a login into a password manager leaves the website’s password unchanged; changing it requires the website’s own account flow.
- Verify a fresh sign-in after an approved password change, and retain access to recovery methods.
- An old connection may still serve a purpose; ask what disconnecting it would break.
- A clean dependency scan covers only part of a system’s security exposure.
Keep the assessment honest
- Yang’s sample report groups findings by area and distinguishes confirmed problems, open work and checks the agent could not complete.
- He frames the tutorial around recent security incidents; the video alone does not independently verify those reports or prove viewers were affected.
- He says 1Password did not sponsor the video; his discussion of its agent platform is explicitly tentative.
“If you can’t check something, say so and give me the exact steps and official link.” — from Yang’s description prompt