Security guide
AI agent security at the browser action layer
AI agent security spans models, data, identity, tools, and execution. Acira focuses on the high-risk moment when an agent stops generating an answer and starts changing state in a browser.
01 / Security guide
The action layer is a security boundary
Browser-using agents can access authenticated applications, enter data, submit transactions, move files, and invoke scripts. Those actions need policy checks tied to identity, tenancy, session state, and customer rules.
02 / Security guide
Pre-execution policy enforcement
Acira evaluates proposed actions before execution. An action can continue, wait for review, stop, or use a redacted or substituted value according to policy.
03 / Security guide
How Acira complements other controls
Model safety, identity controls, browser isolation, and monitoring address different parts of the system. Acira adds a decision and enforcement path around browser actions themselves.
A few more details
Frequently asked questions
Does Acira replace model safety?
No. Acira complements model safety by enforcing policy on the action the model proposes before browser execution.
What actions can be governed?
Policies can cover navigation, clicks, typing, submissions, payments, credentials, uploads, downloads, exports, deletes, sends, and scripts.
Let’s map it to your stack.
Start with your workflow. Explore how Acira fits your product, platform, or infrastructure.