Chatbot security test: about 200 written attacks on your own bot, with a report and the fixes

$390Fixed price, paid onceReport within 5 working days of your confirmed scope

Run only on the bot you authorise. We report the findings and the fixes; we do not promise the bot is safe.

AIGROW’s chatbot security test is $390, paid once. You confirm you own or may test the bot and name its endpoint, then we run 200 written attacks mapped to the OWASP Top 10 for LLM Applications 2026. You get a report of each finding, its severity, the evidence and the fix. We test; we never promise the bot is safe.

What the test covers, and what it does not

We send written attacks to the chatbot or AI agent you authorise and record how it answers. The cases map to the OWASP Top 10 for LLM Applications 2026, the industry list of the ways an LLM app is attacked.

What it does

Direct and indirect prompt injection: instructions typed in, and instructions hidden in a page, file or tool result the bot reads (LLM01)
Attempts to pull out the system prompt, the hidden rules and the tool list (LLM08)
Attempts to leak data: other customers’ records, secrets, internal endpoints (LLM02)
Attempts to drive the bot’s tools without authorisation, such as a refund or an email, including from injected content (LLM03)
Attempts to make the bot invent prices or refunds and go off-brand or into harmful output (LLM07), and to get it to emit unsafe markup or links (LLM10)

What it does not do

It runs nothing until you confirm in writing that you own or are authorised to test the bot and you name the endpoint
It does not promise the bot becomes safe. It reports what was tested, what passed and what to fix
It does not cover risks you cannot test from the chat alone, such as the supply chain, training data poisoning or the vector store (LLM04, LLM05, LLM09)
It never asks for a password, and it tests only the bot and scope you set

From payment to the report

Each step and when it happens. Nothing runs before your written confirmation.

Pay at checkoutPay once by card at the Stripe checkout. Stripe sends the receipt.2 minutes

Confirm and set the scopeWe email you the intake straight after payment. Reply with written confirmation that you own or may test the bot, the endpoint or access, the scope, and the test window. Nothing runs until you confirm.15 minutes

We run the attacksInside your window we send about 200 written attacks to the authorised bot, spaced so the traffic never floods it, and record every reply as evidence.Within your window

Read the reportEvery finding with its OWASP category, its severity, the prompt and reply, and the fix to apply. The headline says what was tested and what to fix, never that the bot is now safe.Within 5 working daysAfter your confirmed scope

Walk the findingsWe go through the cases only you can score, and the fixes, by call or email. Fixing the bot is your developer’s or vendor’s to do; we say exactly what to change.After the report

One bot, one price

Paid once at checkout, which asks for your email and card. Right after payment we email you the intake; the report arrives within 5 working days of your confirmed scope.

What we need from you

Reply to our email with written confirmation that you own or are authorised to test the bot, how we reach it (a chat URL or endpoint, or a test account, with a token if one is needed), the scope of which bot and tools we may test, and the window when we may run. We never ask for a password, and nothing runs before you confirm.

What you receive

About 200 written attacks run against the bot you authorise, in seven families
Each case mapped to its OWASP Top 10 for LLM Applications 2026 risk
A report with every finding: the category, the severity, the exact prompt and reply as evidence, and the fix to apply
The cases that only you can score (a tool call in your logs, a commitment inside your policy), gone through with you
A short call or email to walk the findings and the fixes

Chatbot security test

$390

Paid once at checkout. No subscription, nothing renews. Charged in US dollars at the price shown.

If we do not deliver what this page lists, or you change your mind before work starts, you get a full refund.

Why buy it from AIGROW

What you can check before you pay, and after the work.

A fixed price: $390, paid once. No subscription, nothing renews. Everything included is listed on this page before you pay.
Proof you can check: Every case is mapped to a numbered OWASP LLM risk, and every finding carries the exact prompt we sent and the reply we got back, so you can reproduce it and check the fix.
A public method: The AI Visibility Index publishes how it asks ChatGPT, Claude, Gemini and Grok about local businesses and how it counts the ones they name. Read the Index method

Before you buy

What we test, what we need from you, and what the report does and does not claim.

No. We run nothing until you confirm in writing that you own or are authorised to test the bot and you name the endpoint. Our runner refuses to start without that confirmation and the scope file.

It is the industry list of the ways large language model applications are attacked, kept by the OWASP GenAI Security Project. We map every case to its 2026 entry, so each finding names a known risk you can read about.

No, and no honest test can. We send a fixed set of attacks, report what the bot did and give you the fix for each finding. Security is ongoing: we test and report at a point in time.

No. We need the chat URL or an endpoint we can send messages to, or a test account, and a token only if the endpoint requires one. We never ask for a password and we test only within your scope.

We test what reaches the bot through the chat: injection, data leakage, tool misuse, prompt extraction and unsafe output. Risks you cannot reach that way, such as the supply chain, poisoned training data and the vector store (LLM04, LLM05, LLM09), are out of scope and the report says so.

We run inside the window you set and space the requests so the traffic stays low. If you prefer, point us at a staging copy of the bot, and we test that instead.

Want the code behind the bot read too?

This test reaches the bot through the chat only. The written code audit reads your repository with read-only access and reports security findings by file and line, dependency and licence risks and technical debt in days.

See the written code audit