Skip to main content
A Policy is the active compliance preset on a guard. It tunes the content gate: which patterns to redact, which to block, what audit retention applies, and what permissions are or are not allowed under this regime. The four presets are defined in agentsafe/compliance/presets.py.

​
Choosing a policy

from certior import Guard

guard = Guard(policy="hipaa", budget_cents=5000)
The policy argument accepts a string or a Policy enum value (Policy.DEFAULT, Policy.HIPAA, Policy.SOX, Policy.LEGAL, Policy.LEGAL_PRIVILEGE). Strings are normalized to the enum.

​
The four presets

​
default

The baseline. The content gate runs Certior’s standard scanner; the capability ceiling is unrestricted (max_permissions=["*"]); audit retention is 365 days. Use this for development and for non-regulated production where you still want capability + budget enforcement without strict content rules.

​
hipaa

PHI handling. Activates PII detection and auto-redaction on prompts and tool outputs (names, dates, SSNs, MRNs, addresses, phone numbers). Permission ceiling restricts capabilities the HIPAA preset deems out-of-scope. Audit retention is 2,190 days (6 years).

​
sox

Material non-public information. Activates MNPI keyword and pattern detection; tightens audit with segregation_of_duties=True and 2,555-day (7-year) retention. Use this for any agent that can read or write financial-reporting data. Attorney-client privilege. Blocks routing privileged content across information-flow boundaries, preserves privilege markers, and flags disclosures to non-privileged destinations. Retention is 3,650 days (10 years).

​
What each preset configures

Each ComplianceConfig carries: 
@dataclass
class ComplianceConfig:
    name: str
    content_safety: ContentSafetyPolicy     # the scanner rules
    permissions: list[str]                   # default capability set
    max_permissions: list[str]               # absolute ceiling
    forbidden_permissions: list[str]         # hard deny list
    information_flow_rules: list[FlowRule]   # routing constraints
    audit: AuditConfig                       # retention, tamper-proofness
    required_proofs: list[str]               # obligations Z3 must discharge
When you set policy="hipaa", Guard builds a _safety_policy from that config and the scanner is bound to its rules. The max_permissions ceiling overrides any wider permissions= argument you pass.

​
Custom policies

To extend a preset with domain-specific keywords or rules, build a ContentSafetyPolicy directly and inject it. See agentsafe/safety/scanner.py and agentsafe/compliance/presets.py for the data model. Custom policies are intentionally code-defined rather than free-form YAML.
The content gate is keyword + regex against an auditable rule set - predictable, fast, and never sends content to a third party. It catches the documented patterns exactly; it does not infer violations from paraphrases.

​
See also

  • How it works - the content gate’s place in the pipeline.
  • Capability model - how the permission ceiling interacts with the content gate.
  • Compliance API - listing presets and exporting an execution’s compliance package.