| name | breach-incident-response |
| description | Executes USA breach and security incident response—HIPAA Breach Notification Rule (45 CFR §164.400–414), HITECH 60-day notification, state breach laws, and SOC 2 CC7.4/CC7.5 incident management—for agent, MCP, and LLM-related events. Trigger when investigating suspected PHI/PII exposure, unauthorized MCP access, LLM data leakage, or preparing breach notifications. Do not use for preventive logging design (use audit-logging-integrity) or routine vulnerability scanning. |
Breach Incident Response
Overview
This skill orchestrates security incident and breach response for USA regulatory obligations:
- HIPAA Breach Notification Rule: 45 CFR §164.400–§164.414 (individual, HHS, media notification timelines)
- HITECH: 60-day notification to individuals; HHS OCR reporting
- State breach laws: variable timelines (many 30–90 days)—California, New York, etc.
- SOC 2 CC7.4/CC7.5: Incident identification, response, recovery, communication
AI-specific incidents include: unredacted ePHI sent to non-BAA LLM, MCP tool exfiltration, deanonymization abuse, compromised agent API keys, and prompt injection leading to data disclosure.
When to Use
Use this skill when:
- Suspected or confirmed unauthorized access/disclosure of ePHI, CHD, or confidential data
- LLM data leakage (raw PHI in model logs, training pipeline, vendor breach notification)
- MCP compromise (rogue tool registration, stolen OAuth tokens)
- Agent credential theft or skill tampering detected
- Preparing breach notification letters, HHS OCR reporting, state AG notifications
- Tabletop exercises for agent/MCP incident scenarios
- Post-incident root cause analysis for SOC 2 and HIPAA documentation
Do not use this skill when:
- Designing preventive audit logs (use
audit-logging-integrity)
- General vendor risk questionnaires (use
vendor-third-party-risk)
- Non-security operational outages without data impact (use availability runbooks)
Core Process
Execute steps in order. Contain first—do not delay containment for documentation.
Step 1: Detection and classification
- Triage alert source: SIEM, CCM, user report, vendor notification, red team.
- Classify incident type:
- Security incident (attempted access, no confirmed disclosure)
- Breach (acquisition/access/disclosure of unsecured PHI per §164.402 definitions)
- PCI incident (CHD/SAD suspected compromise)
- Assign severity and incident commander.
- Start immutable incident timeline (UTC timestamps, actor actions).
Step 2: Immediate containment
- Isolate affected systems without destroying forensic evidence:
- Revoke compromised MCP OAuth tokens and API keys
- Disable affected agent deployments; block LLM API keys
- Snapshot logs and MCP server state before changes
- Preserve evidence with SHA-256 hashes; chain of custody log.
- Do not deanonymize logs for investigation unless authorized forensic path.
Step 3: Scope and impact assessment
- Determine what data was involved:
- ePHI elements, volume, individuals affected
- CHD/SAD if payment systems involved
- Determine unsecured PHI analysis per §164.402:
- Was data encrypted per HIPAA guidance such that disclosure is impracticable?
- Risk assessment per four-factor test (§164.402(2)): nature of PHI, unauthorized person, acquisition/view, mitigation
- Use audit logs (
audit-logging-integrity artifacts) for timeline reconstruction.
- Document minimum necessary findings—avoid spreading PHI in incident tickets.
Step 4: HIPAA breach determination
- If unsecured PHI involved, apply breach presumption (§164.402)—breach unless low probability of compromise documented.
- Privacy officer/legal confirms breach vs security incident.
- If breach: start notification clock (60 days to individuals per §164.404; HHS per §164.408).
Step 5: Notification workflows
HIPAA (if breach confirmed)
| Audience | Requirement | Timeline |
|---|
| Individuals | Written notice (mail or email if opted) | Without unreasonable delay, max 60 days |
| HHS OCR | Portal submission | ≤60 days; <500 individuals = annual log |
| Media | If >500 residents of a state | Without unreasonable delay, max 60 days |
- Draft notification content per §164.404(c): description, types of PHI, steps for individuals, CE contact, mitigation steps.
- Do not include unnecessary PHI in notification letters.
State laws
- Identify affected individuals' states; check state breach statutes (timeline may be shorter than HIPAA).
- Notify state AG where required (e.g., CA, NY thresholds).
SOC 2 CC7.5
- Document customer notification if incident affects service commitments or confidentiality.
- Align with contractual incident notification SLAs.
Step 6: PCI incident response (if CHD involved)
- Engage PCI forensics (PFI) if required by acquirer/brands.
- Do not store full PAN in incident docs—use tokenized references.
- Cross-notify payment brands per merchant agreement.
Step 7: Eradication and recovery
- Remove attacker access: rotate all secrets, patch vulnerabilities, remove rogue MCP tools.
- Restore from known-good baselines; re-run
compliance-as-code-governance scans.
- Verify redaction gate and MCP allowlists before re-enabling agent production traffic.
Step 8: Post-incident review
- Root cause analysis: technical, process, human factors.
- Update skills, runbooks, CCM rules to prevent recurrence.
- Document lessons learned for board/management (CC1.2, CC2.2).
- Retain incident records per HIPAA documentation requirements (6 years common).
Step 9: Regulatory reporting completion
- Track notification deliveries and HHS submission confirmation.
- Archive all artifacts in encrypted evidence store with access controls.
Common Rationalizations
| Excuse the agent might generate | Required rebuttal |
|---|
| "It's probably not a breach—skip notification planning." | HIPAA applies breach presumption until documented four-factor risk assessment proves low probability. |
| "We contained it—notification can wait." | Notification clocks start upon discovery—delay beyond regulatory timelines is a secondary violation. |
| "Put full incident details in Slack for speed." | Incident channels must use approved secure tools—no cleartext PHI in Slack unless BAA-covered and scoped. |
| "LLM vendor deleted the data—no breach." | Vendor assurance does not replace your breach determination and notification obligations. |
| "Only 10 patients—no HHS report needed." | HHS reporting still required (annual log if <500)—individual notification still required. |
| "Forensics can wait until after we're back online." | Snapshot logs and MCP state before recovery actions destroy evidence. |
Red Flags
- Raw ePHI confirmed in non-BAA LLM vendor logs
- Deanonymization tool used anomalously without ticket correlation
- Incident commander lacks privacy/legal engagement on PHI breach
- Notification letters drafted past 60-day HIPAA deadline
- Full patient list posted in unencrypted incident spreadsheet
- MCP rogue tool remained registered during containment phase
- No HHS OCR submission tracking for reportable breach
Verification