{
  "id": 7405498,
  "title": "Abnormal AI: Amazon Bedrock AgentCore for agentic email security at scale",
  "url": "https://urgent.news/2026/09/14/abnormal-ai-amazon-bedrock-agentcore-for-agentic-email-security-at",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-14T21:22:45.000Z",
  "source": {
    "name": "AWS Machine Learning",
    "slug": "aws-machine-learning",
    "url": "https://aws.amazon.com/blogs/machine-learning/abnormal-ai-amazon-bedrock-agentcore-for-agentic-email-security-at-scale/"
  },
  "original_language": "en",
  "account": "Amazon Bedrock AgentCore Code Interpreter is a fully managed, serverless runtime designed to enable agents to execute code dynamically at scale. This capability is crucial for tasks that go beyond semantic reasoning, such as basic math, data processing, and code verification. Abnormal AI, a security service protecting over 25% of Fortune 500 companies, utilizes this feature to support its real-time inline email threat detection.\n\nAbnormal AI's architecture comprises three tiers of detection: high-volume classification (Tier 1), machine learning models for uncertain cases (Tier 2), and inline agents that use Code Interpreter for inline threat detection (Tier 3). The inline agents analyze threat intelligence data within a sandbox environment, write scripts dynamically, and evaluate the data's relevance to the overall behavioral model.\n\nAbnormal AI's use of AgentCore Code Interpreter reflects their AI-native approach to building production-level systems. Their sandbox approach provides a secure environment for code execution, with configurable time-to-live, security measures, flexible networking options, and limited file handling. They have also integrated preloaded runtimes with common libraries for visualization, statistics, and data processing.\n\nA key aspect of their system is the analyst agent, which operates in batch mode. This agent ingests misclassifications and tuning signals from the real-time pipeline, identifies patterns, and autonomously writes draft candidate heuristics for Tier 1 and improved models for Tier 2. This continuous learning process helps improve the overall accuracy of their email threat detection systems.",
  "summary": "Learn how Abnormal AI deployed Amazon Bedrock AgentCore Code Interpreter as an ephemeral compute scratch pad for the agents behind its real-time email threat detection at billion-message scale, plus the sandbox design decisions and practical lessons for builders deploying Code Interpreter in production.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 3,
    "also_reported_by": [
      {
        "outlet": "AWS Machine Learning",
        "title": "How Ninth Wave built AI-powered open finance onboarding on Amazon Bedrock",
        "url": "https://urgent.news/2026/09/14/how-ninth-wave-built-ai-powered-open-finance-onboarding-on-amazon",
        "published": "2026-09-14T15:58:57.000Z"
      },
      {
        "outlet": "Dev.to",
        "title": "GigWatch: AI-powered remote job monitoring (open source)",
        "url": "https://urgent.news/2026/09/14/gigwatch-ai-powered-remote-job-monitoring-open-source",
        "published": "2026-09-14T19:51:29.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}