Glossary
AI governance, term by term.
The vocabulary of this field does a lot of quiet work in vendor decks, so here it is with the mystery removed. Every definition stands on its own, whether you write policy, review security, or just want to know what a tool call actually is.
AI governanceAI governance is the practice of deciding what AI tools and agents may do, enforcing those rules while the AI acts, and keeping proof of what happened.Security as controlSecurity as control is the practice of deciding what an AI may do before it acts, rather than detecting and reporting what it did afterwards.Shadow AIShadow AI is AI use that happens outside an organization's approved tools and accounts, through personal logins and unvetted apps, where no policy applies and no record is kept.
Terms
Seventeen terms, defined in plain language
Each card carries its own link so you can point a colleague straight at one, and each term hands off to the page where the depth lives.
- AI governance
- AI governance is the practice of deciding what AI tools and agents may do, enforcing those rules while the AI acts, and keeping proof of what happened.
- How Verillian does it
- Security as control
- Security as control is the practice of deciding what an AI may do before it acts, rather than detecting and reporting what it did afterwards.
- How Verillian decides
- Shadow AI
- Shadow AI is AI use that happens outside an organization's approved tools and accounts, through personal logins and unvetted apps, where no policy applies and no record is kept.
- Control it, don't just find it
- Audit trail
- An audit trail is a chronological record of the actions taken in a system, kept so that what happened can be reconstructed and checked later.
- The whole-turn record
- Tamper-evident
- Tamper-evident means a record is written so that any later change to it becomes visible, which lets a reviewer test the record instead of trusting it.
- Why it counts as evidence
- Hash chain
- A hash chain links records by giving each entry a cryptographic fingerprint of the one before it, so altering any entry visibly breaks the chain from that point on.
- The chain, explained
- Endpoint
- An endpoint is a device an organization manages, such as a laptop or workstation, and it is where AI use actually happens, before traffic is encrypted to a provider.
- Why the device is the place
- Agent
- An agent is an AI system that takes actions on its own, reading files, running commands, and calling tools in pursuit of a goal rather than just answering a question.
- Governing agents
- Tool call
- A tool call is the moment an AI stops talking and does something, invoking a capability like running a command or writing to a system, and it is the unit an AI policy rules on.
- Ruling on tool calls
- MCP
- MCP, the Model Context Protocol, is an open standard that lets AI agents connect to outside tools and data sources, such as a CRM or a code repository.
- MCP under policy
- Redaction
- Redaction replaces sensitive values in a text, such as a patient identifier or an account number, with placeholders before that text is shown or sent anywhere.
- Redaction, proven in the record
- Deny by default
- Deny by default is the rule that anything not explicitly allowed is refused, so a gap in policy fails safe instead of open.
- The rule above the rules
- Fail-closed
- Fail-closed describes a control that stops traffic whenever it cannot do its own job, so an error never quietly becomes an open door.
- Fail-closed by construction
- Air gap
- An air gap is the complete physical separation of a network from the internet, used where data is too sensitive to risk any outside connection at all.
- Running air-gapped
- AI kill switch
- An AI kill switch is a single control that halts AI use across an organization at once, for the moments when something needs to stop before it is understood.
- How stopping works
- Checkpoint
- The checkpoint is the part of Verillian that runs on each device, reading AI traffic where it is still readable and applying declared policy to every request and tool call.
- The layer end to end
- Admin server
- The admin server is the management server a Verillian customer runs on their own infrastructure, never on Verillian's. It distributes signed policy, tracks enrolled devices, and stores the sealed record under the customer's key alongside the searchable metadata.
- Where your data lives
Words defined.
Now see them in action.
A tool call refused, a record sealed, a switch flipped: the terms that matter most on this page show up live in the demo, on infrastructure like yours.