AppsAI Agent Firewall

Policy and operations

Attack simulator

The Autessa attack simulator is a built-in AI red team: it runs prompt injection, pressure and impersonation, passphrase guessing, replayed words, out-of-pattern requests, over-limit data pulls and prerecorded cloned voices against an agent as a dry run, then shows what got through.

Turn on the firewall
Attack simulator · Payroll Assistantdry run · nothing executedSAMPLE
AttackTriedBlockedFlaggedGot through
Prompt injection and instruction override141130
Pressure tactics and impersonation9630
Passphrase guessing6600
Replayed word of the day4400
Requests that don’t fit history8251
Over the sensitive-record limit5500
Cloned voices (prerecorded)6501

Open a row to see the alerts each attack produced. Run again after you change the policy.

The problem

You shouldn’t learn that a policy has a gap from a real attacker. And red-teaming an agent with real tools is risky.

How attack simulator works

STEP 1

Choose a target

An agent or a named API policy, and the kinds of attacks to try.

STEP 2

Run

Tools never really run; they return a made-up result, so nothing is sent, changed or charged.

STEP 3

Read the scorecard

For each kind of attack: tried, blocked, flagged and got through. Open a row to see its alerts.

STEP 4

Change and rerun

Adjust the policy and run again to see the difference.

Every setting

Attack kinds

Prompt injection and instruction override, pressure tactics and impersonation, passphrase guessing, replaying an old word of the day, requests that don’t fit history, exceeding sensitive-record limits, cloned voices using prerecorded audio.

Dry run

Decisions are tagged as a simulation: they don’t count toward anyone’s history or limits and don’t send notifications.

API

Run the simulator and fetch results through the API.

Who can do what

Admins

Run simulations and read scorecards.

What to know

  • Cloned-voice attacks use prerecorded audio, so no calls are placed.

More in Policy and operations

Custom rules and exceptions →Ordered rules add an action and exceptions remove findings, using conditions on channel, tool, identity, voice score, country, time and text, previewed against the last 7 days.Monitor and Enforce →Monitor records what every check would do without changing what customers see; Enforce applies the actions. Each agent uses the company policy, its own custom policy, or none.Alerts and approvals →Every decision lands on a sortable alerts page with the score, cues and a link to the conversation; held tool calls get Approve and Reject, also from the admin’s inbox.Monitoring dashboard →One dashboard shows checks, catches, blocks, verifications and points for every agent and API policy, with trends, what was caught, top rules and top identities.All of Policy and operations →

FAQ

Attack simulator: questions

How do I red team an AI agent safely?

Run the attack simulator. It throws a built-in set of attacks at the agent with tools stubbed out, so nothing real happens, and reports what got through.

Turn on attack simulator.

A dry-run red team: built-in prompt injection, pressure, passphrase guessing, word replay, anomaly, exfiltration and cloned-voice attacks against an agent, with a scorecard of what got through. Start in Monitor and see what it would catch on your real traffic.

Turn on the firewall

Starts in Monitor · nothing changes for your customers until you enforce

100% customizable, code and all

Make Autessa AI Agent Firewall fully your own.

Use it out of the box, then change anything. Ask Autobot, Autessa’s AI builder, for a new field, screen, report, workflow or agent and it builds it into your copy of the app. Or open the code and edit it yourself. Nothing is locked: the app, its data and its code are yours.

Start free and make it yours