Cantina Trust, Governance & Auditability

Cantina Trust, Governance & Auditability is a component of the Cantina security platform that provides governance controls, transparency mechanisms, and audit logging for AI-driven security agents that perform autonomous remediation actions in production environments. The platform is designed around three core concerns: what actions an agent is permitted to take, how much autonomy it operates with, and what it has done. These concerns are addressed through configurable autonomy settings, enumerated tool scopes, and full audit trails. Autonomy controls: - Agents can be configured to act fully autonomously or to pause and await human approval before executing consequential actions - Signal rule changes (edits, disabling, deletions) require explicit approval before taking effect - Escalated issues can be investigated by an agent, which then recommends a verdict, but a human confirms the close - Code fixes are proposed as pull requests and wait for human review by default; agents can be granted merge permissions explicitly Scope controls: - Investigation tools and action tools are separated by type; read-access tools cannot write - Each skill (agent capability) has an enumerated list of permitted actions - The ceiling of what an agent can do is fixed by the skill definition, regardless of autonomy level granted Transparency and auditability: - Skills are versioned, human-readable documents showing investigation steps, decision thresholds, and callable tools - Signal rules are inspectable down to their exact match logic, conditions, and time windows - Agent reasoning is recorded with each verdict, including confidence levels and unverified items - Every action is attributed to the agent that took it and the approval it ran under - Dismissed issues retain the reasoning used to dismiss them Platform security: - The team behind Cantina conducts vulnerability research and public disclosures - A Trust Center is available for security documentation