Policy library
Policy templates for AI agents
10 free, CC0 starting points for agent controls around personal data, health data, financial records and bulk actions. Each one has the policy JSON and YAML, example tests and a link that opens it in Vulnify in monitor mode.
Block bulk deletes
Blocks any delete that touches more than 100 records in one action, so a looping agent cannot wipe a table.
Block bulk exports of personal data
Blocks any export of more than 1,000 records from resources you classify as customer personal data (CUSTOMER_PII) in one action. A starting point for internal controls around large personal-data exports. Vulnify’s risk engine may also return REVIEW or BLOCK for bulk personal-data exports on its own, even without this policy (for example, large internal exports), so not every REVIEW or BLOCK on these exports comes from this template.
Block bulk external exports of personal data
Blocks exports of more than 1,000 records from resources you classify as customer personal data (CUSTOMER_PII) sent to an external destination. Vulnify’s risk engine may also return REVIEW or BLOCK for bulk personal-data exports on its own, even without this policy (for example, large internal exports), so not every REVIEW or BLOCK on these exports comes from this template.
Block financial data sent outside
Blocks any agent action that sends data from resources you classify as FINANCIAL to an external destination.
Block sensitive (e.g. health) data leaving the company
Blocks any agent action that sends data from resources you classify as SENSITIVE to an external email or API. The SENSITIVE classification alone is enough to match.
Review agent access to sensitive (e.g. health) data
Every agent read of a resource you classify as SENSITIVE waits for approval, whether or not that particular read contains health data: the SENSITIVE classification alone is enough to match. Ships in monitor mode first: decisions are recorded but the action is allowed, so you can see volume before enforcing.
Review changes to financial records
Any agent write to a financial resource (ledgers, payouts, refunds) waits for an owner or admin to approve. Vulnify does not read amounts, so this covers every write, not a value threshold.
Review deletes and exports outside business hours
Deletes and exports outside Monday to Friday, 09:00 to 18:00 in your org time zone wait for approval. Ships in monitor mode.
Review emails with sensitive data to external addresses
Emails an agent sends to an external address that are flagged as containing sensitive data need approval before they go out. The flag can come from the resource classification or from sensitive data Vulnify detects in the email content (for example CPF numbers or email addresses).
Review personal data sent outside the company
Any agent action that sends data from resources you classify as customer personal data (CUSTOMER_PII) to an external email or API waits for a human approval. Related to LGPD art. 46. Does not determine whether a destination is outside Brazil; international transfers have separate requirements under LGPD arts. 33–36.
Not legal advice. These templates are starting points to help you design internal controls for AI agents. They are not legal advice and do not make your organization compliant with LGPD, GDPR, HIPAA or any other law or regulation. Review them with your own legal and compliance team before relying on them. Templates are provided “as is”, without warranty. Whether a rule matches depends on how you classify your resources and on the data your agents send; test in monitor mode before enforcing.
License: the templates are dedicated to the public domain under CC0 1.0. “Vulnify” is a trademark; the CC0 dedication does not cover the Vulnify name or logo.