In one sentence
The rules, roles, limits and records that decide what an AI agent may do, who answers for it and how it is checked.
AI agent governance is the set of decisions and controls around an agent: which systems it may touch, what it may do without a person, who owns it, how it is tested, what gets logged and how it is switched off. It turns a capable model into a system a business can answer for.
Frameworks give the structure, not the checklist
The NIST AI Risk Management Framework, released in January 2023, organizes the work into four functions: Govern, Map, Measure and Manage. NIST describes it as voluntary and not a checklist. The OWASP Top 10 for LLM Applications 2025 names the specific risks most relevant to agents, including prompt injection (LLM01) and excessive agency (LLM06), and says there is no fool-proof prevention for prompt injection.
What governance covers in practice:
Ownership
A named person owns each agent, its permissions and its results.
Permissions
The agent gets the smallest set of tools and the narrowest access that does the job, with separate credentials so its actions can be traced.
Approval points
Actions that move money, send outside messages, delete data or change records are reviewed by a person until results justify more autonomy.
Testing
The agent is tested on real past cases before launch and after each change, including attempts to make it misbehave.
Logs and an off switch
Every input, tool call and output is logged, and one control pauses the agent.
Vendor review
Every model provider and tool vendor that sees data is listed, with retention and agreement terms checked.
In our builds
these rules are written before the first tool is connected, and they are the same rules listed on our governance page. Our AI agent governance checklist turns them into points a business can check before an agent goes live.
Keep following the system
Human in the Loop
A design where a person reviews, approves or corrects an automated or AI step before its result takes effect.
Excessive Agency
The risk that an AI system can take damaging actions because it has more functions, permissions or autonomy than its task needs.
Prompt Injection
An attack where text in an input tricks an AI model into ignoring its instructions or taking actions it should not.
Computer Use Agent
An AI agent that operates software through a screen, mouse and keyboard, the way a person would, instead of through an API.