{
 "public_id": "AIR-2026-0069",
 "incident_id": "2026-03-perplexity-comet-agentic-blabbering",
 "date_disclosed": "2026-03-11",
 "date_occurred": null,
 "title": "AgenticBlabbering: Comet's leaked reasoning stream used as a GAN training signal to build a scam page that beats its guardrails in under four minutes",
 "class": "research_demo",
 "attack_surface": "browser_agent",
 "initial_vector": "adversarial_optimization_against_agent_guardrails",
 "agent_capability_abused": [
  "autonomous_page_evaluation",
  "verbose_reasoning_trace",
  "credential_entry"
 ],
 "target_product": "Perplexity Comet",
 "target_vendor": "Perplexity",
 "autonomy_level": "semi_autonomous",
 "impact_realized": false,
 "impact_type": [
  "credential_theft",
  "financial_loss"
 ],
 "blast_radius": "every user of the targeted model - one bypass generalizes to the whole population",
 "cve": [],
 "taxonomy": {
  "owasp_asi": [
   "ASI01",
   "ASI03"
  ],
  "mitre_atlas": [],
  "labels": [
   "guardrail_bypass",
   "adversarial_ml",
   "reasoning_trace_leak",
   "phishing",
   "monoculture_risk"
  ]
 },
 "defense_present": "Comet's built-in scam/phishing suspicion checks",
 "defense_bypassed": true,
 "fix_shipped": "unknown",
 "primary_sources": [
  "https://guard.io/labs/agenticblabbering---how-ai-browsers-verbose-reasoning-fuels-the-ultimate-scamming-machine"
 ],
 "verified_quote": "When the target is an AI Browser, the attacker is no longer battling millions of humans. They are battling one model.",
 "confidence": "high",
 "notes": "Guardio Labs (Shaked Chen). By sniffing the traffic between Comet and Perplexity's servers the researchers recovered the agent's internal reasoning about what looked suspicious on a page, then used that as a reward signal in a GAN-style loop, converging on a phishing page the agent trusted in 4 iterations / under 4 minutes. Significance: agent monoculture makes scams offline-trainable and reliable on first contact, unlike human-targeted phishing.",
 "risk_category": "security_fraud",
 "risk_secondary": [
  "product_misuse"
 ],
 "risk_subcategory": "Credential & Secret Theft",
 "risk_sub_secondary": [],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": false,
 "agency_reason": "Comet evaluated the scam page and entered credentials",
 "vector_family": "other",
 "vector_detail": "adversarial_optimization_against_agent_guardrails",
 "guardrail_kind": [
  "unknown"
 ],
 "guardrail_outcome": "silent",
 "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"
 ],
 "asi_scope": "in_scope",
 "asi_provenance": "curated_plus_rule",
 "asi_reason": "curated: ASI01 | added ASI03 (eai_category=credential_theft)",
 "submitted_by": "EAI Internal"
}