CERT-AGID observed a phishing campaign abusing the INPS name, logo, and visual identity to lure victims to a fake site that collects personal documents and photos. The fraudulent portal also uses an AI-based file validation step, reportedly identified as Gemini 1.5 PRO, to filter uploads and increase the quality of stolen data. #INPS #CERT-AGID #Gemini1.5PRO
Keypoints
- The campaign uses fake SMS messages that impersonate INPS and direct users to a counterfeit website.
- The phishing pages closely mimic official institutional portals and guide victims through a five-step data submission flow.
- Attackers request highly sensitive information, including personal details, identity documents, pay slips, CUD forms, and selfies.
- A new feature in this wave is an AI-based validation layer that checks uploaded files before accepting them.
- The system can semantically describe image content, reject clearly irrelevant pictures, and appears to rely on a vision model rather than simple heuristic checks.
- The same validation can still accept invalid documents, including a fake ID card marked āFACSIMILE,ā showing inconsistent filtering.
- CERT-AGID activated takedown procedures for malicious domains and shared relevant IoCs through its feed to accredited organizations.
MITRE Techniques
- [T1566.001] Phishing: Spearphishing Link ā Victims receive fraudulent SMS messages containing a link to a fake INPS website designed to steal data (āGli SMS fraudolenti invitano lāutenteā¦ con un link che conduce a un sito contraffattoā).
- [T1204.001] User Execution: Malicious Link ā The attack depends on the user clicking the SMS link and following the guided upload flow on the fraudulent portal (āinvitano lāutente a procedere al controllo dei propri datiā).
- [T1587.001] Develop Capabilities: Malware ā The phishing kit includes an AI-based validation component to inspect uploaded files and classify their contents (ālāintroduzione di un ulteriore livello di controlloā¦ basato su un modello di intelligenza artificialeā).
- [T1056.002] Input Capture: GUI Input Capture ā The fake site collects personal data through a structured multi-step form, including identity documents and selfies (āfinalizzato al caricamentoā¦ dettagli anagrafici, documenti identificativi, buste paga, CUD e selfieā).
- [T1113] Screen Capture ā The operator leverages uploaded images of identity documents and other documents as visual evidence to validate and harvest sensitive material (ācaricamento della carta di identitĆ ā).
- [T1027] Obfuscated Files or Information ā The fake validation layer hides its true logic behind AI-generated outcomes and structured error messages to conceal automated checks (āvalutazione semanticaā¦ campo pubblico errorā).
- [T1112] Modify Registry ā Not mentioned in the article.
Indicators of Compromise
- [Domain / URL] Fraudulent phishing site and upload endpoint used by the campaign ā upload_ajax.php, and the fake INPS domain referenced in the SMS link.
- [File / Document content] Uploaded or requested identity and payroll documents ā carta di identitĆ , buste paga, CUD, and selfie.
- [String / UI text] Fraudulent validation and error messages shown to victims ā āVerifica AI in corsoā¦ā, āRicarica una foto piĆ¹ chiara e leggibileā, and āFACSIMILEā.
- [HTTP response fields] AI validation response structure observed during uploads ā ok:true, motivo:āā, and ai_rifiuto:true.
- [Service / Model identifier] Self-identified AI component used by the fake portal ā Gemini 1.5 PRO.
- [IOC feed reference] CERT-AGID campaign indicators distributed to accredited organizations ā IoC feed, and related malicious domains.