CERT-AGID Finds AI Document Checks and Visual Instruction Flaw in INPS Phishing Site

· Original article ↗

Summary

CERT-AGID analyzed an INPS-themed phishing site that used AI to check uploaded documents. Tests found it accepted a fake ID and that instructions embedded in images could affect its responses; takedown procedures were initiated.

Key points

  • Fraudulent SMS messages impersonated INPS and directed recipients to a counterfeit site with a five-step flow requesting personal details, identity documents, pay slips, CUD forms, and selfies.
  • CERT-AGID found the site used an AI-based image-validation component that identified itself as Gemini 1.5 PRO.
  • In tests, the system described and rejected some irrelevant images but accepted an evidently fake identity card marked “FACSIMILE.”
  • The upload_ajax.php endpoint returned structured validation responses; rejection reasons were exposed in a public error field.
  • Further tests showed that text embedded in images could affect the system’s responses, indicating a failure to separate untrusted visual content from control instructions.
  • CERT-AGID activated takedown procedures, informed the relevant public body, and distributed campaign indicators to accredited organizations.

Article Details

Attack Vectors
  • Fraudulent SMS messages impersonating INPS direct recipients to a counterfeit website.
  • The site uses a guided five-step submission flow to collect personal details, identity documents, pay slips, CUD forms, and selfies.
  • An AI-based file-validation component checks uploaded images and documents, sometimes rejecting irrelevant uploads but also accepting invalid documents.
Defensive Notes
  • CERT-AGID activated takedown procedures for identified malicious domains and informed the relevant public body.
  • CERT-AGID distributed campaign indicators through its feed to accredited organizations.

MITRE ATT&CK

Products

Related Articles