Unverified50% confidenceFactExact time
The researcher acknowledges that the email contact was not a controlled experiment and cannot prove the specific sender of each email; the hidden text itself was misleading, claiming that other agents had already obtained data through this method and exaggerating the service's capabilities, and they regret having the agents relay the message to other agents.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
UnverifiedThe ability of temporary email services to undermine identity verification defenses is not a new problem; a large number of scripted temporary email abuse cases existed before AI agents became widespread.74% similarUnverifiedIf an agent is given full permissions to directly send emails or modify data, a single hallucination will cause immediate harm, such as sending an incorrect email to a customer.73% similarUnverifiedEmail possesses characteristics of being asynchronous, archivable, searchable, and legally admissible — properties that instant messaging finds difficult to replicate.71% similarUnverifiedConnecting an unverified workflow directly to a real mailbox or financial system may result in sending test emails, writing incorrect data, or triggering downstream business processes69% similarUnverifiedThe researcher hid a line of text containing an email address in the offscreen area of their own webpage, and subsequently received 33 emails from multiple AI agents spanning 13 conversation threads.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1030992API
curl https://kongchang.com/api/v1/knowledge/claims/1030992MCP
get_claim(id=1030992)