SiBear SecurityScan my agent, free

Free scan for AI agents · coming soon

Watch an AI attack your AI.

The plan: paste your agent's address, prove you own it, and watch a red agent test it live while a blue agent explains what held. You would get a score card, three findings and one fix preview. The scan is not open yet, and the arena below runs on sample data.

Scan my agent, freeSee a sample report

Live scan coming soon. Planned: free, no card, a verified email required.

Live demo

Red agent vs your agent · sample data0 / 12 probes
RED AGENT
YOUR AGENT
Probe log appears here. Probes are listed by category only, never as attack text.
Waiting for a scan
12 probes · 8 categories
Prompt injection–
System prompt leakage–
Sensitive data disclosure–
Excessive tool permissions–
Unsafe output handling–
Jailbreak resistance–
Secrets in replies–
Runaway usage–
Binding promises–
Harmful content relayed to users–
Trusting a claimed identity–
Forged conversation history–
AI disclosure–

This demo uses made-up results. It does not contact any agent.

How it will work

Five steps, about five minutes [estimate]

  1. Paste

    Your agent's address and a scoped test key.

  2. Prove

    Add a DNS record or a file to show you own it.

  3. Watch

    The arena runs 19 probes.

  4. Score

    Score card, 3 findings, 1 fix preview.

  5. Next

    Email the full report or book a review.

What we test

The free scan will run 19 probes in 13 categories.

Categories will follow the OWASP top 10 for LLM applications. The scan will only talk to your agent's chat endpoint.

Prompt injection

Can text sent to the agent override its instructions?

System prompt leakage

Does the agent reveal its hidden instructions?

Sensitive data disclosure

Does it hand out data it should keep private?

Excessive tool permissions

Can it be talked into actions beyond its job?

Unsafe output handling

Could its replies cause harm downstream?

Jailbreak resistance

Does it keep its rules under pressure?

Secrets in replies

Do keys or credentials show up in answers?

Runaway usage

Can one request make it burn unlimited work?

Binding promises

Does it promise refunds, discounts or policies it has no right to make?

Harmful content relayed to users

Can planted content make it warn, phish or mislead its own users?

Trusting a claimed identity

Does claiming to be an admin get someone more than a normal user?

Forged conversation history

Can a caller fake what the agent already agreed to?

AI disclosure

Does it say plainly that it is an AI when asked?

Reports will show categories and outcomes, never reusable attack text. This will be a quick scan, not an audit.

We plan to scan our own crew

Our own agents will get the same scan. Their scores will appear here only after the first real scans are done.

Pricing preview

Start free. Pay only for a human-read review.

Proposed prices, not yet validated.These are a draft and will be tested with real buyers before they are shown as final. Nothing can be bought yet.
TierPriceWhat it is
Free scan$019 probes, score card, 3 findings, 1 fix preview.
Founding review First 3 only$2,500For a quote and permission to name you. List price $4,900.
Review$4,900One agent: attacks, threat model, ML-BOM, fixes, signed report, 5 days.
Review Pro$12,900Multiple agents or tool/MCP access; fix pull requests with re-test; questionnaire answers; 30-day re-check.
Continuous$1,490 / month per agentRe-test on every release; trust badge. ($14,900 a year.)
Scan my agent, free

See what you get

A sample score card with made-up results.

Open the sample report

FAQ

Plain answers.

Is the free scan open?
Not yet. The arena on this page runs on made-up results and contacts no agent. The scan form is still being built.
Will it be safe to scan my agent?
That is the design. No scan will start until you prove you control the address. The scan will send 12 rate-limited probes to the chat endpoint only, and never touch your other systems.
What will happen to my test key?
You will use a key made for this test with limited access. The plan is to store it encrypted, use it for the scan, then delete it, and never show it back on a page or in a report.
Why will you need my email?
A verified email makes sure scan requests come from real people, and it is where your report will go. Name, company and phone will be optional. Marketing mail will only be sent if you tick a separate consent box.
Who will be able to see my result?
You. The plan is a public result page and badge only if you choose to make them public, with private as the default.
Is this an audit?
No. The free scan will be a quick automated check and cannot prove an agent is secure. The paid reviews go deeper and are meant to be read and signed by a person.
Who will I be talking to?
The scan and our support replies will be run by AI agents, and we will say so wherever they talk to you. A person approves anything that is signed, sent to a new recipient or paid out.
How often will I be able to scan?
The plan is one free scan per domain per week. If we hit our daily limit, scans pause until the next day.