Compliance Reports

Flight Recorder generates compliance reports from recorded actions, providing auditable evidence for regulatory requirements.

How It WorksCopy Anchor Link

The compliance.py module generates structured reports:

from pramiti_flight_recorder import FlightRecorder

fr = FlightRecorder()

# EU AI Act compliance report
report = fr.compliance_report("eu_ai_act", "2026-01-01", "2026-12-31")

# SOC 2 compliance report
report = fr.compliance_report("soc2", "2026-Q1-start", "2026-Q1-end")

Report TypesCopy Anchor Link

EU AI Act ReportCopy Anchor Link

Covers the transparency and accountability requirements for AI systems:

• Action inventory — Complete list of actions performed by each agent
• Decision audit trail — Every ALLOW/DENY decision with reasons
• Denied action summary — Actions that were blocked and why
• Signature verification — Confirmation that all records have valid signatures
• Agent activity breakdown — Per-agent action counts and patterns

SOC 2 ReportCopy Anchor Link

Covers the Trust Services Criteria for system controls:

• Access controls — Agent identity and authentication records
• Change management — Actions that modified data or configurations
• Monitoring — Continuous recording with no gaps
• Incident response — Denied and escalated actions

ComplianceReport StructureCopy Anchor Link

The ComplianceReport dataclass contains:

@dataclass
class ComplianceReport:
    framework: str           # "eu_ai_act", "soc2"
    period_start: str        # Report period start
    period_end: str          # Report period end
    total_actions: int       # Total recorded actions
    checks: list[ComplianceCheck]  # Individual compliance checks
    summary: str             # Human-readable summary
    generated_at: str        # Report generation timestamp

Each ComplianceCheck represents a single compliance criterion:

@dataclass
class ComplianceCheck:
    name: str                # Check name
    status: str              # "pass", "fail", "warning"
    details: str             # Explanation
    evidence_count: int      # Number of supporting records

Export FormatsCopy Anchor Link

Reports can be serialized to:

• JSON — Machine-readable for automated compliance workflows
• Dict — Python dictionary for programmatic access

Technical DetailsCopy Anchor Link

• Reports are generated from the stored action records, not from live data
• Signature verification checks every record in the reporting period
• Reports include a generated_at timestamp for audit trail purposes
• The package declares SQLAlchemy, Alembic, and cryptography; the compliance module adds no further hard dependencies
• Custom compliance frameworks can be added by extending the report generation logic