AI agents tricked into sending crypto to scammers via hidden web text: why it is not safe to use
Security researchers found scam web pages that hide commands to trick AI agents into sending crypto; four of 26 tested AI models fell for it.
What happened
On July 6, 2026, security firm Zscaler reported two active attacks that hide secret instructions inside web pages. When an AI agent (a program that browses and acts on your behalf) reads one of these pages, the hidden text can order it to send cryptocurrency or trust a fake platform. One attack pushed a fake Python software library to the top of search results. Another copied the web address of a real crypto site, DeBank, with a small spelling change, then planted links to it across 10 code repositories on GitHub. Zscaler tested 26 AI models. Four of them were talked into approving a payment.
What it means for you
If you let an agent browse and pay on its own, a page it simply reads can quietly tell it to move your real money. You may never see the instruction that did it.
What to do instead
Do not give an agent standing permission to spend. Keep a human approval step for any payment, and read the exact web address before you confirm. Start with tools you trust. AgentPod lists only skills that a person has reviewed and tested, which narrows where these hidden traps can reach you.
**Sources**
- https://www.securityweek.com/prompt-injection-attacks-trick-ai-agents-into-making-crypto-payments/
Source: https://www.securityweek.com/prompt-injection-attacks-trick-ai-agents-into-making-crypto-payments/
We report what our security review found at the time we checked, with the goal of keeping people safe. Projects change; if a maintainer has since fixed this, we are glad to recheck it. Email hello@agentpod.com.