Nixer audits AI agents, LLMs, code repos, CI/CD pipelines, and cloud configs for the exposures HIPAA/HITRUST/SOC2 auditors are starting to ask about — and most health tech teams have never tested.
Three attack surfaces that go undetected in traditional HIPAA assessments — and show up in every Nixer scan of a health tech stack.
HIPAA's Security Rule was written for structured EHR databases, not LLM APIs with unrestricted context windows. Your clinical AI now processes unstructured notes, lab results, and patient history — none of which maps cleanly to §164.312 controls. Auditors are starting to ask, and most teams don't have answers.
prompt-injection-unvalidated-inputIndirect prompt injection — where malicious instructions hidden in patient records, referral notes, or lab attachments redirect your LLM's output — is the most underdetected threat in health tech AI right now. One poisoned intake form can instruct your agent to exfiltrate records to an attacker-controlled endpoint. Nixer detects injection-vulnerable input pipelines before they reach your models.
indirect-prompt-injectionCrewAI, LangChain, and LlamaIndex agents are being deployed with EHR tool access, scheduling APIs, and internal knowledge bases. These frameworks have documented CVEs — sandbox escapes, deserialization vulnerabilities, tool-call poisoning — that create direct paths to patient data. Nixer maps your agent dependency tree and surfaces exploitable paths before your next audit.
agent-framework-cveA single scan run covers the full attack surface your clinical AI touches — not just the model layer.
Agent definitions, system prompts, tool registrations, and MCP server configs scanned for injection vectors, over-privileged tool access, and unsafe output routing.
✓ Covers PHI agent accessInput sanitization gaps, prompt construction patterns, and context window injection paths in the application code that calls your clinical LLM APIs.
✓ Catches injection pathsSource code scanned for hardcoded credentials, insecure dependencies, CVE-tracked frameworks, and secrets in version control — the most common HIPAA audit finding.
✓ No secrets in codeGitHub Actions, GitLab CI, and build workflows audited for workflow injection, unpinned actions, and GITHUB_TOKEN over-exposure — attack chains that reach production PHI.
✓ Supply chain checksLLM-assisted analysis of your system design for mismatched trust boundaries, missing access controls between agent layers, and PHI flow paths that bypass audit logging.
✓ Trust boundary analysisCloud configs and IAM policies reviewed for over-privileged agent service accounts, misconfigured PHI storage buckets, and cross-account access patterns.
✓ IAM blast radiusEvery finding Nixer surfaces maps directly to the controls auditors check. Here's what lands in your audit report.
| Nixer Finding | HIPAA Security Rule | HITRUST Control | SOC2 Criteria |
|---|---|---|---|
|
Prompt Injection in Clinical LLM prompt-injection-unvalidated-input |
§164.312(a)(1) — Access Control | Control 09.ab — Monitoring System Use | CC6.1 — Logical and Physical Access |
|
PHI Exfiltration via Tool Call agent-tool-call-data-exfil |
§164.312(e)(1) — Transmission Security | Control 09.s — Information Leakage | CC6.7 — Restricts Transmission of ePHI |
|
Insecure Agent Framework (CVE) agent-framework-cve |
§164.308(a)(5) — Security Awareness | Control 10.m — Control of Technical Vulnerabilities | CC7.1 — Detects and Monitors for Vulnerabilities |
|
Workflow Injection in CI/CD pwn-request-workflow |
§164.312(b) — Audit Controls | Control 09.aa — Audit Logging | CC8.1 — Change Management |
|
Hardcoded Credentials / Secrets hardcoded-secrets |
§164.308(a)(3) — Workforce Access Management | Control 01.a — Access Control Policy | CC6.3 — Access Based on Least Privilege |
Four free modules on prompt injection, tool abuse, RAG poisoning, and CI/CD pipeline security. Built for developers and security engineers shipping AI in regulated industries.
Book a 20-minute demo and we'll walk through a live scan of your AI stack — or a comparable health tech repo — and show you exactly what auditors will find.
"Health Tech AI Security Checklist" — 12 items every CISO should verify before the next audit.