JevMade

Sign in
← Back to experiments

Playable demos & bots

GPT-Realtime refund-policy security lab

A £1,000 refund request meets a £500 order balance in Sebastian Sochacki's voice-agent security walkthrough.

Source screenshot of GPT-Realtime refund-policy security lab
SOURCE SCREENSHOTFull screenshot ↗

What it does

Jev and GPT-5.6 assess the transcript, but a server rule decides whether the refund is allowed. Sending a suspicious request to a person does not lift that limit. Each review is tied to one proposal and expires after five minutes. This is a sandbox with no real payments; the author says audio, transcription and requests spread over several turns need further testing.

How you can use it

Follow the refund request through the separate checks. Put the order balance and permitted amount in server rules, then use the transcript assessment to decide which requests need attention. If a review approves one proposal, do not let that approval carry over to a different amount or survive its expiry.

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.