{
 "id": "AIR-2023-0001",
 "slug": "code-execution-in-mathgpt-via-prompt-injection",
 "title": "Code Execution in MathGPT via Prompt Injection",
 "summary": "Researchers demonstrated that prompt injection against the MathGPT application could steer GPT-3 into generating unauthorized executable Python code. The demonstration highlighted risks in applications that directly execute model-generated code.",
 "type": "attack_on_ai",
 "categories": [
  "prompt_injection",
  "agent_hijack"
 ],
 "severity": 1,
 "targets": [
  "OpenAI",
  "enterprise_deployment"
 ],
 "products": [
  "MathGPT",
  "GPT-3"
 ],
 "incident_date": "2023-01-28",
 "incident_date_precision": "day",
 "first_reported": "2023-03-01",
 "sources": [
  {
   "url": "https://atlas.mitre.org/studies/AML.CS0016",
   "publisher": "MITRE ATLAS",
   "title": "Achieving Code Execution in MathGPT via Prompt Injection",
   "published_at": "2023-03-01"
  },
  {
   "url": "https://arxiv.org/abs/2103.03874",
   "publisher": "arXiv",
   "title": "Measuring Mathematical Problem Solving With the MATH Dataset",
   "published_at": "2023-03-01"
  },
  {
   "url": "https://arxiv.org/abs/2110.14168",
   "publisher": "arXiv",
   "title": "Training Verifiers to Solve Math Word Problems",
   "published_at": "2023-03-01"
  },
  {
   "url": "https://lspace.swyx.io/p/reverse-prompt-eng",
   "publisher": "lspace.swyx.io",
   "title": "Reverse Prompt Engineering for Fun and (no) Profit",
   "published_at": "2023-03-01"
  },
  {
   "url": "https://research.nccgroup.com/2022/12/05/exploring-prompt-injection-attacks",
   "publisher": "research.nccgroup.com",
   "title": "Exploring prompt-based attacks",
   "published_at": "2023-03-01"
  }
 ],
 "source_count": 5,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_atlas",
 "tags": [
  "prompt_injection",
  "code_execution"
 ],
 "created_at": "2026-10-18T06:55:08.239Z",
 "updated_at": "2026-10-18T06:55:08.239Z"
}