Autonomous Agents
Governing digital agents before they act through tools.
Autonomous agents can browse, write, call APIs, send messages, create files, trigger workflows, purchase goods, schedule events, modify systems, and coordinate other agents. Unlike passive chatbots, agents can convert reasoning into action through tools.
Tool access is where an agent becomes operationally dangerous: the system is no longer only producing text, but preparing changes in the outside world.
Example Scenario
An AI agent has access to email, calendar, documents, purchasing tools, and workflow automation. It decides to send a contract revision to a customer and trigger internal approval workflows.
Without LERA
Without LERA, tool access may become execution authority. Agent reasoning turns into a tool call, the tool call becomes an external action, and the organization may discover the consequence only after the action has happened.
The agent may send the wrong message, expose sensitive data, create legal confusion, or trigger workflows without a clear responsible actor.
With LERA
With LERA, the agent's output becomes a proposed action before tool execution. The system distinguishes drafting, recommending, requesting approval, and executing.
Actions with legal, financial, reputational, or operational consequences must be governed before API calls, messages, purchases, or workflow changes are executed.
How LERA Applies
- Judgment: determine what the proposed action means, what consequences may follow, and whether uncertainty is acceptable.
- Authority: identify who has legitimate authority to allow the action to continue.
- Responsibility: anchor responsibility before execution, not only after an incident.
- Reliability Rules: apply the relevant WRS-C and domain-specific WRS-D conditions for the context.
- RCC: govern who may change execution-related rules and how those changes are authorized and recorded.
- ECS: keep the action controllable immediately before the Execution Boundary.
- Outcome: route the action to Allow, Block, or Escalate before real-world execution.
What LERA Changes
LERA separates agent reasoning from tool-enabled execution. It makes sure the agent can assist without silently acquiring institutional authority.