Caging the Agents: A Zero Trust Security Architecture for Autonomous AI in Healthcare

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• The proposed security architecture employs a six-domain threat model for agentic AI in healthcare, covering credential exposure, execution capability abuse, network egress exfiltration, prompt integrity failures, database access risks, and fleet configuration drift.method 70%
  Evidencepartial

  We develop a six-domain threat model for agentic AI in healthcare covering credential exposure, execution capability abuse, network egress exfiltration, prompt integrity failures, database access risks, and fleet configuration drift.

  Implicationpartial

  Explicitly stated in the abstract.

  Verificationpartial

  partial

• The security architecture implements a four-layer defense in depth strategy including kernel-level workload isolation using gVisor on Kubernetes, credential proxy sidecars, network egress policies, and a prompt integrity framework.method 70%
  Evidencepartial

  We implement four-layer defense in depth: (1) kernel level workload isolation using gVisor on Kubernetes, (2) credential proxy sidecars preventing agent containers from accessing raw secrets, (3) network egress policies restricting each agent to allowlisted destinations, and (4) a prompt integrity framework with structured metadata envelopes and untrusted content labeling.

  Implicationpartial

  Explicitly stated in the abstract.

  Verificationpartial

  partial

• The architecture was deployed for nine autonomous AI agents in production at a healthcare technology company.technical 70%
  Evidencepartial

  This paper presents a security architecture deployed for nine autonomous AI agents in production at a healthcare technology company.

  Implicationpartial

  Explicitly stated in the abstract.

  Verificationpartial

  partial

• During a 90-day deployment period, four HIGH severity security findings were discovered and remediated by an automated security audit agent.result 70%
  Evidencepartial

  We report results from 90 days of deployment including four HIGH severity findings discovered and remediated by an automated security audit agent

  Implicationpartial

  Explicitly stated in the abstract and supported by the analysis section 'method_eval'.

  Verificationpartial

  partial

• The deployment demonstrated progressive fleet hardening across three VM image generations.result 70%
  Evidencepartial

  progressive fleet hardening across three VM image generations

  Implicationpartial

  Explicitly stated in the abstract and supported by the analysis section 'method_eval'.

  Verificationpartial

  partial

• The defense coverage of the architecture maps to all eleven attack patterns identified in recent literature.result 70%
  Evidencepartial

  defense coverage mapped to all eleven attack patterns from recent literature.

  Implicationpartial

  Explicitly stated in the abstract and supported by the analysis section 'method_eval'.

  Verificationpartial

  partial

• The effectiveness of the architecture might be limited by its configuration complexity and the necessity for constant updates to address emerging threats.limitation 70%
  Evidencepartial

  The architecture's effectiveness might be limited by its configuration complexity and the necessity for constant updates to address emerging threats.

  Implicationpartial

  Stated as a caveat in the analysis section.

  Verificationpartial

  partial

• The proposed security framework has the potential to replace existing, less specialized security solutions in the healthcare sector for AI agent deployment.market 70%
  Evidencepartial

  This security framework can potentially replace existing, less specialized security solutions in the healthcare sector, providing a more tailored and robust defense for AI agent deployment.

  Implicationpartial

  Implied by the 'disruption' section of the analysis, suggesting it offers a more tailored and robust defense.

  Verificationpartial

  partial

Startup potential card