Attestorian- AI agent
- Software that uses an AI model to decide on and take actions, such as calling tools, APIs, or other systems.
- Tool call
- A request from an AI agent to perform an action, for example retrieving data, updating a record, or creating a payment.
- Evidence package
- A portable, sealed file (.attpkg) containing the recorded steps of an AI session, which any standalone verifier can check offline.
- Seal
- A digital signature over the chained record of steps. Any alteration to any step makes verification fail.
- Tamper-evident
- A property of a cryptographic record where any unauthorized edit, deletion, reordering, or insertion is deterministically detectable by verification.
- Hash (digital fingerprint)
- A fixed-length cryptographic value computed from data (such as SHA-256). Changing even one character of the input data produces a completely different value.
- Selective redaction
- Removing specific sensitive fields from a shared copy of an evidence package while keeping the cryptographic root seal verifiable.
- Prompt injection
- Hidden instructions placed inside untrusted content, such as an uploaded invoice, document, or email, designed to manipulate an AI agent into unauthorized actions.
- Four-eyes or M-of-N approval
- A pre-execution governance rule requiring M out of N authorized humans to co-approve a high-impact action before it runs, binding their identities into the sealed record.
- Shadow mode
- Recording and sealing AI evidence asynchronously from existing telemetry traces without sitting in the live execution path.
- Inline guard mode
- Sitting directly on the AI agent tool-call path so policies, parameter bounds, and human approvals are enforced before actions execute.
- Counterfactual replay
- Re-running a recorded AI session offline under a stricter or modified policy to see exactly which steps would have been paused or blocked.
- Offline verification
- Checking an evidence package on your own machine without a network connection or an account with the evidence producer.
- KMS and HSM
- Key Management Service and Hardware Security Module: dedicated enterprise systems that generate, store, and protect cryptographic signing keys.
- MCP (Model Context Protocol)
- An open standard for connecting AI agents to external tools, APIs, and data sources.
- OpenTelemetry
- An open standard and observability framework for collecting traces, metrics, and telemetry from software systems.
- Lineage
- The verifiable record of which AI model, prompt template, tool definition, and software version produced a given action.
- Row-level security
- Access control rules enforced at the data layer so each tenant, business unit, or reviewer only accesses records authorized for their scope.
Explore AI agent security architecture.
Review our glossary and technical documentation with your engineering team. The USD 10,000 pilot fee is credited in full.
Last updated: