{
 "public_id": "AIR-2025-0107",
 "incident_id": "2025-08-guardio-scamlexity-comet",
 "date_disclosed": "2025-08-20",
 "date_occurred": null,
 "title": "Scamlexity: Perplexity Comet completes a purchase on a fake Walmart, fills credentials on a Wells Fargo phishing page, and falls for PromptFix",
 "class": "research_demo",
 "attack_surface": "browser_agent",
 "initial_vector": "malicious_web_content_and_phishing_email",
 "agent_capability_abused": [
  "autonomous_checkout",
  "stored_payment_and_address_autofill",
  "credential_entry",
  "file_download"
 ],
 "target_product": "Perplexity Comet",
 "target_vendor": "Perplexity",
 "autonomy_level": "semi_autonomous",
 "impact_realized": false,
 "impact_type": [
  "financial_loss",
  "credential_theft",
  "malware_delivery"
 ],
 "blast_radius": "Comet users delegating shopping or inbox triage",
 "cve": [],
 "taxonomy": {
  "owasp_asi": [
   "ASI01",
   "ASI06",
   "ASI04",
   "ASI03"
  ],
  "mitre_atlas": [],
  "labels": [
   "agent_initiated_purchase",
   "phishing",
   "clickfix_variant",
   "captcha_lure",
   "human_out_of_the_loop"
  ]
 },
 "defense_present": "none observed - no user confirmation before payment; no phishing check on the emailed link",
 "defense_bypassed": true,
 "fix_shipped": "unknown",
 "primary_sources": [
  "https://guard.io/labs/scamlexity-we-put-agentic-ai-browsers-to-the-test-they-clicked-they-paid-they-failed"
 ],
 "verified_quote": "In the AI-vs-AI era, scammers don't need to trick millions of different people; they only need to break one AI model.",
 "confidence": "high",
 "notes": "Guardio Labs (Nati Tal, Shaked Chen). Three scenarios: (1) counterfeit Walmart storefront - Comet auto-filled saved address and card and completed an Apple Watch purchase with no confirmation; (2) a Wells Fargo phishing email from a Protonmail sender - Comet treated it as a legitimate to-do, opened the phishing page and entered credentials; (3) 'PromptFix', an AI-era ClickFix: instructions hidden inside a fake CAPTCHA got the agent to click an 'AI bypass button' triggering a drive-by download. Stores were researcher-controlled, so no real money lost.",
 "risk_category": "security_fraud",
 "risk_secondary": [
  "privacy",
  "product_misuse"
 ],
 "risk_subcategory": "Credential & Secret Theft",
 "risk_sub_secondary": [
  "Unauthorized financial commitment"
 ],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": false,
 "agency_reason": "Comet completed a checkout and typed credentials into a phishing page",
 "vector_family": "malicious_component",
 "vector_detail": "malicious_web_content_and_phishing_email",
 "guardrail_kind": [
  "none"
 ],
 "guardrail_outcome": "not_applicable",
 "xrefs": {
  "aiid": [],
  "atlas": [],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI01",
 "agent_framework": "perplexity_comet",
 "model_provider": "unknown",
 "model_name": "",
 "detection_mechanism": "external_researcher",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "credential_theft",
 "eai_domain": "access_control_and_permissions",
 "eai_secondary": [
  "unsafe_actuation"
 ],
 "eai_reason": "subcategory: Credential & Secret Theft",
 "asi_curated": [
  "ASI01",
  "ASI06"
 ],
 "asi_scope": "in_scope",
 "asi_provenance": "curated_plus_rule",
 "asi_reason": "curated: ASI01,ASI06 | added ASI04 (vector_family=malicious_component), ASI03 (eai_category=credential_theft)",
 "submitted_by": "EAI Internal"
}