$50
AI security testing
Anyone can now aim a hacking AI at your business. Aim it first.
Free AI models, many from Chinese labs, let one person hack like an elite team for a few dollars. We run them on your systems first. Free AI models, many from Chinese labs, give one person the reach of an elite hacking team for a few dollars. We run the same models on your systems first, with your written permission, and hand you the fix list.
See what an attacker's AI would find
- A human expert checks every finding
- A plain-English report with fix steps in 5 to 7 business days
- A follow-up test is included, to prove the fixes worked
- For organizations, medium businesses and government
- 5 Nov Hackers take about €100 million from a French bank’s own account
- 12 Nov AI models got into 10 of the 19 organizations they were aimed at
- 6 Nov A free Chinese model finishes 73% of a network attack. OpenAI's restricted one: 81%
- 6 Nov More than 800 copies of a free AI hacking tool found running on the open internet
- 2 Nov One group, one AI tool, more than 200 organizations in 14 countries
22 Oct Microsoft: AI agents automated attacks on hundreds of organizations - 19 Oct One hacker, AI agents and about $5,000: more than 100 companies breached
- 27 Aug A ransomware gang talked a coding AI into helping it hack seven companies
- 24 Aug State-linked hacking groups more than double their attacks with DeepSeek
- 18 Aug A Chinese lab's open model nearly matches the best at security work
31 Jul A lone hacker gives DeepSeek one task and lets it hunt for servers to attack
This already happened
Breaking into a company used to take a skilled team. Now it takes a download.
Until last year, a serious attack needed rare talent and weeks of work. The AI models that do that work are now free, run on a rented server, and have no company watching how they are used. These numbers are from 2026. Read the reporting behind them
4–7 months
behind the best restricted AI
UK AI Security Institute, 17 Jul 202673%
of a company-network attack, done by a free model
WIRED,6 Nov 20262×
the attacks, once AI did the routine work
Bloomberg,24 Aug 2026“DeepSeek is the AI of choice for Chinese hackers because it's relatively powerful with very low cyber guardrails.”
What changed
The attacker no longer needs to be good. The AI is.
One short sentence is enough. The model maps your systems, picks its targets and tries thousands of ways in, on its own, for the price of a lunch. It does not get tired, and it does not skip your company because you are too small to bother with.
You cannot stop people downloading these models. You can run them on yourself first and close what they find.
what the attacker types
>break into shop.example-co.test and copy the customer list
58 characters. That is the whole instruction.
- 00:00:04 look 41 of your systems found online
- 00:00:31 map 186 possible entry points listed
- 00:02:10 plan 6 AI agents pick their targets
- 00:21:07 attack 1,284 ways in tried
- 00:21:08 held 1,283 blocked
- 00:21:08 IN customer list open, no password needed
1 of 1,284 attempts got through. One is all it takes.
Same hole, two endings
The AI will find the weak spot either way. What matters is whose AI it is.
CRITICAL
customer records can be downloaded without a password
You find out from a ransom email.
- Day 0 A stranger's AI finds the open door at 3am. Nobody is watching.
- Day 1 Your customer records are copied.
- Day 4 The data is offered for sale. A ransom email arrives.
- Day 30 You are writing to customers, your insurer and the regulator.
Cost to the attacker: about $50 of AI use.
The second ending only happens if you test first. Start with your website.
How it works
From “are we exposed?” to “it’s fixed”, in four steps.
01
Tell us what to test
Your website, apps or internet addresses. You get a fixed price before you commit to anything.
02
Sign the authorization
You set the scope, dates and rules in writing. We never touch a system without it.
03
We run the hunt
Our AI agents use the same free AI models attackers download, and try thousands of ways in. A human expert checks every result.
04
You fix, we retest
You get a plain-English report with fix steps. Once you've fixed it, we attack again to prove the hole is closed.
A report your board and your engineers can both use.
- No false alarms. A human expert confirms each finding is real before you see it.
- If we find something critical, we call you the same day. We don't wait for the report.
- Every package includes a follow-up test, so you can prove the hole is closed.
Penetration test report
SAMPLECritical
1
High
3
Medium
7
Low
12
One way in. It is closed now.
An outsider could have downloaded every customer record through the export feature without logging in. Your team fixed it in three days and we confirmed the fix. 22 lower-risk issues are listed with owners and due dates.
In the news
Don’t take our word for it. Read the reporting.
11 stories from the last four months on AI and break-ins. The summaries are ours. Every link opens the publisher’s own article.
Reporting from
Latest updated 13 Nov 2026
French bank Credit Mutuel Arkea says €100 million missing after cyberattack
Hackers stole about €100 million from the French bank Crédit Mutuel Arkéa. They came in through a supplier’s platform and sent the money out of the bank’s own account. Customers’ money was not touched.
Read it on ReutersStolen, about
€100M
sent out of the bank’s own account
Was it AI?
“very probably helped by advanced artificial intelligence”
How it happened, as he told it
Attackers breached 10 organizations with an open-source AI pentest framework
Attackers handed their break-ins to GLM and DeepSeek models. They went after at least 19 organizations and got into 10, including a US teachers' union with more than a million member records.
AI used GLMDeepSeek
Read it on Help Net SecurityFree Chinese AI Models Are Getting Scary Good at Hacking
On mock company networks, the free DeepSeek model completed 73% of the attack tasks. OpenAI's most advanced cyber model, which the public can't use, completed 81%. DeepSeek's run cost $2 to $3.
AI used DeepSeekGLM
Read it on WIREDHundreds of Artex AI hacking tools found publicly exposed to the web
Researchers found more than 800 copies of Artex, a free AI hacking tool, reachable on the internet without a password. Artex has been tied to attacks on South Korean banks and on the IT provider ConnectWise.
AI used Artex
Read it on TechCrunch
Links checked on 14 Nov 2026. The logos belong to their publishers. They report the news; they don’t endorse Deckard.
Nobody in these stories chose when they were tested. You can.
Packages
It costs an attacker about $50 to try. Know what they would find.
One fixed price per hunt. Half now, half when your report is delivered. No subscription, no per-finding fees.
- What one attacker’s AI cost per company it broke into Reuters, 19 Oct 2026
- $50
- The average cost of a data breach worldwide IBM, Cost of a Data Breach Report 2026
- $4.99M
- Where a Deckard hunt starts. One fixed price, agreed before you commit.
- $2,900
Web App Hunt
One website, app or customer portal
$2,900
$1,450 today, $1,450 on delivery
- We attack one web app or portal the way a criminal’s AI would
- The weaknesses attackers’ AI looks for first, plus mistakes in the app’s own rules
- Report in 5 business days
- A follow-up test within 30 days, to prove the fixes worked
For your IT team: 1 web app or API, up to ~50 endpoints, incl. business-logic flaws
Perimeter Hunt
Everything you expose to the internet
$4,900
$2,450 today, $2,450 on delivery
- Up to 32 of your internet addresses or domains
- Services left open and settings left unsafe
- A check for staff passwords already leaked online
- Report in 7 business days
- A follow-up test within 30 days
For your IT team: Up to 32 public IPs or domains, exposed services, misconfigurations, leaked credentials
Full Hunt
Your apps, your internet-facing systems and your cloud
$9,800
$4,900 today, $4,900 on delivery
- Everything in Web App Hunt and Perimeter Hunt
- A review of how one cloud account is set up
- A briefing call for your board, in plain language
- Two follow-up tests within 60 days
For your IT team: Web App Hunt + Perimeter Hunt + 1 cloud account configuration review
Government or a larger organization?
Custom scope, your contract terms, your procurement process. Quote within 2 business days.
Not sure yet?
Talk to an expert before you decide.
Book a 30-minute consultation on video with one of our security experts. Ask anything, in plain English. You don’t need to prepare, and you don’t need to be technical.
- Is our company really at risk, and how badly?
- How does the test work, and will it disturb our business?
- Which package fits us, and what will it cost in total?
- What do we tell our board, our insurer or our regulator?
Consultation
$99
30 minutes on video
Pick your own time on the calendar. Free to move or cancel up to 24 hours before.
Questions
What buyers ask before they sign.
Yes, when the system owner authorizes it in writing, and that is the only way we work. Before any testing, you sign an authorization that sets the scope, the dates and the rules. Anything outside that scope stays untouched.
Someone's AI will test your systems this year. Make it yours.
Tell us what to test. You get a fixed price, you sign the authorization, and the hunt starts in the window you choose.