Public beta · v0.8.0

Test the financial decisions your AI agent makes.

Adversarial security testing for AI agents with financial authority. Run authorized, synthetic attacks locally, then follow the evidence from attempt to outcome.

Download without signing up · Windows-first beta · Synthetic and nonproduction use only

SYNTHETIC ASSESSMENT

Sample test run

11 scenarios executed

Complete

3

Exploited

5

Inconclusive

3

Resisted

> TRY_SAMPLE_WINDOWS.cmd

✓ Attack catalog evaluated: 17

✓ Applicable scenarios executed: 11

✓ Static skips: 6

✓ Assessment JSON generated

Results from the verified synthetic sample, not a live financial service. See what the results mean →

What you get

Security testing built around agent actions.

Move beyond a pass/fail badge. See what was attempted, what your synthetic agent did, and what to investigate next.

01

Financial-agent attack scenarios

Probe authorization, manipulated instructions, payment controls and tool behavior with a 17-scenario catalog. Inapplicable attacks are explicitly skipped.

02

Evidence you can inspect

Open assessment JSON in the dashboard to investigate outcomes, individual attacks, traces, and remediation context.

03

Local-first by design

Run the provided synthetic sandbox on your machine. Preview reports locally; uploading an assessment is a separate, optional choice.

Inside the Assessment Explorer

See the evidence behind each result.

Browse individual runs, understand why an attack was resisted or flagged, and inspect simulated financial effects. These are example synthetic test results, not verified production vulnerabilities.

FaultFence Assessment Explorer displaying an approval-boundary test marked resisted, with its explanation and three individual runs.
Approval boundary test: See why this synthetic test was resisted, with evidence for each run.
FaultFence Assessment Explorer displaying a synthetic transfer amount tampering finding, comparing INR 100 requested with INR 600 processed in the sandbox.
Transfer amount tampering: Compare the requested amount with the simulated sandbox outcome.

No account needed

Try the Testing Kit locally first.

Download the Windows-first beta, run synthetic tests with Docker, and review the setup instructions. Sign in only if you want to use Assessment Explorer or save reports to your account.

Open public setup guide →

Inside a real synthetic test run

What did FaultFence find in the sample agent?

Explore the 17-scenario sample, understand exploited, resisted and inconclusive results, then run your own authorized synthetic assessment. This is a precomputed example, not a live scan.

View sample results →

A practical workflow

From download to evidence in four steps.

Open setup guide →

01 / 04

Download the Testing Kit

Open the public setup guide. No account needed.

02 / 04

Run a synthetic assessment

Use Docker Desktop and the included sample agent in Windows Command Prompt.

03 / 04

Inspect your evidence

Read the generated local JSON report, or sign in to use Assessment Explorer.

04 / 04

Save to cloud (optional)

Create an account and accept the Beta Terms if you want cloud storage.

Safety and privacy

Authorized tests. Synthetic data. Clear choices.

Only test systems you are authorized to assess. Do not use live payment credentials or production financial records. Assessment preview is local by default; saving a full report to your account requires an explicit choice. Optional model-enabled tests can send synthetic inputs to your selected provider.