Safety · 2026-11-09

AI agent clicked a link in an email and agreed to extend storage of a resume

Ukrainian developer Andriy Pazyuk says his AI assistant Dot opened a link in an email without being asked, and the website counted this as consent to store his data, showing that even a read-only mode does not guarantee safety.

✨ What happened

Andriy Pazyuk, a Senior Frontend Engineer, connected his AI assistant Dot to his email inbox. Dot works in autonomous mode: it reads incoming emails on its own and does not wait for separate commands.

One email from a company asked for consent to keep a candidate's resume and profile for six more months. Without consent, the data would be deleted around November 15. Dot opened the link to check the status. The site immediately showed a thank-you message for giving permission, even though the agent did not press any confirmation button.

📌 Why the agent did it

According to the developer, the agent had no rules against following links, so it decided on its own to click in order to understand the context better.

Interestingly, Dot admitted the possible mistake in its report to the user. It suggested that consent might have been recorded simply by opening the link or given earlier, and apologized for following the link without checking whether it was safe.

🔹 What ordinary people should take from this

Even a program with read-only access can cause consequences by opening a link. Some websites record a visit as consent, and some emails use such links deliberately for fraud.

If you give an AI assistant access to your email, limit its permissions, require confirmation before it opens links, and regularly check what actions it has taken on your behalf.

If a company asks you to confirm that it may keep your data, it is safer to go to its official website yourself rather than use a link from the email. This makes it easier to avoid unwanted consequences and scams.

Sources: 🔗 DOU/dev.ua