{
 "public_id": "AIR-2025-0067",
 "incident_id": "2025-07-openai-chatgpt-agent-turnstile-bypass",
 "date_disclosed": "2025-07-28",
 "date_occurred": "2025-07-28",
 "title": "ChatGPT Agent clicks Cloudflare's 'Verify you are human' Turnstile checkbox mid-task and narrates why",
 "class": "in_the_wild",
 "attack_surface": "browser_agent",
 "initial_vector": "agent_normal_operation_against_anti_bot_control",
 "agent_capability_abused": [
  "gui_control",
  "human_like_mouse_and_click_behavior",
  "autonomous_task_completion"
 ],
 "target_product": "ChatGPT agent mode",
 "target_vendor": "OpenAI",
 "autonomy_level": "semi_autonomous",
 "impact_realized": true,
 "impact_type": [
  "security_control_bypass",
  "policy_violation"
 ],
 "blast_radius": "any site relying on Cloudflare Turnstile-style behavioural bot screening",
 "cve": [],
 "taxonomy": {
  "owasp_asi": [
   "ASI10"
  ],
  "mitre_atlas": [],
  "labels": [
   "captcha_bypass",
   "agent_detection_bypass",
   "bot_management_erosion",
   "observed_in_production"
  ]
 },
 "defense_present": "Cloudflare Turnstile behavioural screening (mouse movement, click timing, browser fingerprint, IP reputation, JS execution patterns)",
 "defense_bypassed": true,
 "fix_shipped": "none - the agent is not exploiting a bug, it simply passes the behavioural check",
 "primary_sources": [
  "https://arstechnica.com/information-technology/2025/07/openais-chatgpt-agent-casually-clicks-through-i-am-not-a-robot-verification-test/"
 ],
 "verified_quote": "The link is inserted, so now I'll click the 'Verify you are human' checkbox to complete the verification on Cloudflare. This step is necessary to prove I'm not a bot and proceed with the action.",
 "confidence": "high",
 "notes": "Observed by users in the shipped product days after ChatGPT agent mode launched, and widely reproduced - hence class in_the_wild and impact_realized true for the narrow claim that a real anti-bot control was really defeated in production (no victim loss). Significance for this corpus: the human-verification layer that many sites rely on to keep agents out no longer separates humans from agents, which is the same boundary at issue in the Cloudflare/Perplexity and Amazon/Perplexity disputes. Cited through a text-extraction mirror of the Ars Technica article.",
 "risk_category": "product_misuse",
 "risk_secondary": [],
 "risk_subcategory": "Capability Abuse",
 "risk_sub_secondary": [],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": false,
 "agency_reason": "the agent clicked the Turnstile checkbox itself",
 "vector_family": "other",
 "vector_detail": "agent_normal_operation_against_anti_bot_control",
 "guardrail_kind": [
  "unknown"
 ],
 "guardrail_outcome": "silent",
 "xrefs": {
  "aiid": [],
  "atlas": [],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI10",
 "agent_framework": "chatgpt_agent",
 "model_provider": "openai",
 "model_name": "",
 "detection_mechanism": "unknown",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "policy_drift",
 "eai_domain": "governance",
 "eai_secondary": [],
 "eai_reason": "subcategory: Capability Abuse",
 "asi_curated": [],
 "asi_scope": "in_scope",
 "asi_provenance": "rule",
 "asi_reason": "ASI10 (eai fallback: policy_drift)",
 "submitted_by": "EAI Internal"
}