Hackers now send AI agents to break into websites. Is yours next?
Full-spectrum protection and threat intelligence against hostile AI agents: see what they would find, catch them in the act, know exactly who they are, and shut them out of every site in the network. Without ever hiding you from ChatGPT or Google.
Free. Reads only what any visitor can see. Your score is saved so you can share it; nothing sensitive is ever shown publicly.
This month, every protected site · updated hourly
...
attacks
...
AI agents
...
stolen keys used
How it stops them
1 Scan: see what a hostile AI agent finds first on your site, free.
2 Catch: decoys trap it in the act: fake keys, files and logins that report back when touched.
3 Identify: script, person or AI agent, how sure, how far it got, whose network, where to report it.
4 Shut out: caught on one site, refused on every site, with an alert to you. Google and ChatGPT are never blocked.
Built by Kyle Clouthier. I set up the decoys, set the headers and clean up leaks for small business and agency sites, and I answer the chat myself. Ontario, Canada.
A hacker no longer has to sit at a keyboard. They point an AI agent at thousands of sites, and it hunts for leaked passwords and keys, tries them, and keeps going while its owner sleeps. In October one operator did this to more than 1,200 websites and got into more than 100 companies, for about $50 a break-in. One leaked key file is all it takes: a fake AWS key stolen from our own decoys was used to try for admin rights within three hours.
The free scan reads only what any visitor can see and shows what a hostile agent would find first: leaked keys and files, open forms, missing locks, and hidden text that can hijack AI agents.
Website safety · Prompt injection · Honeypot · News · Plans
Agencies, IT and multi-site owners
Scan all your sites at once
Paste up to ten sites and get one list, worst first: the score, the platform and the first thing to fix on each. Then get the list by email with one price for the weak ones.
More than ten, or client sites you look after every week? Agency scans 25 every Monday with reports under your own name, and the prospect finder does 100 at a time.
Case file · 24 October 2026 · from our own logs
A fake key, taken and used within three hours
Our decoy files hold a fake AWS key that reports back the moment anyone tries it. This is what happened the first day it was live. Times are UTC. Nothing real was at risk; on a real site, the last entry is a takeover.
03:04 UTC · secondstrike.io
A scanner on a Google Cloud address asks for 40 fake files in two seconds: .env variants, .git, backups, and /.aws/credentials. It now has the fake AWS key.
04:04 UTC · simgen.dev
A second scanner, same tool, starts on simgen.dev: 126 fake files over four hours, including the fake AWS key.
05:43 UTC · AWSstolen key used, 2 h 39 min after the theft
The first scanner tries the key against AWS: lists servers, lists regions, checks whether the key can send email (the spam test), reads who the key belongs to.
06:42 UTC · simgen.dev
An address on Tencent Cloud, with a fake iPhone browser string, follows the hidden link that robots.txt forbids.
07:43 UTC · simgen.devAI agent
A second Tencent address, same browser string, reads the note meant only for AI agents and calls the check endpoint. Sixty-one minutes after the link, from a new address.
10:26 UTC · AWStakeover attempt
The second scanner makes 50 calls in one minute with the key: read secrets, create an access key, create a bucket, stop and reboot servers, attach admin policies to its user and to a role.
Both scanners ran the same tool, Amazon's own .NET SDK, from rented cloud servers. The AI agent is a separate visitor: one that read a note no person sees and acted on it. The same decoys are now a free plugin.
In the news · updated 23 Oct 2026, 22:26 UTC
AI agents and websites: what is happening
- 23 OctA developer’s post went viral (about 663,000 views): put a flag.txt file in your site’s root with a prompt injection that tells a breaking-in AI agent it passed the test, then to patch its... @Nexuist on X
- 21 OctFortune: experts say more capable AI models are behind a growing wave of attacks. Georgetown CSET’s Jessica Ji says AI systems are now more capable of hacking and act more autonomously; the... Fortune
- 20 OctA hacker used autonomous AI agents, run on the open-source tool Cairn, to target more than 1,200 websites and break into more than 100 companies, including Deere, SharkNinja and Cummins.... The Hacker News
Live decoys
Right now, attackers and their AI agents are hunting websites for passwords
We leave fake password files, fake keys and a fake admin login on our own sites and on every site running Guard. No real visitor ever touches them, so everything that does is an attacker: scanners, people, and more and more AI agents. These are this month's real numbers. Search engines are verified and never see a decoy.
The network
Every site that joins makes every other site safer and smarter
Threat intelligence from live decoys on every site in the network. When one site catches an attacker using a stolen key, every other site refuses it within the hour. The more sites join, the less room hostile agents have to move.
Shared protection
One site catches an attacker. Every site locks it out.
When an address uses a stolen decoy key or the decoy admin password on any Guard site, its logins, sign-ups and form posts are refused on every Guard site within the hour. It can still read pages, so nothing breaks for real visitors, and every listing ends 14 days after the last sighting.
Loading the live list...
Totals only are public, in the Agent Attack Index. Addresses are never published. Caught by mistake? Ask to be removed.
Rather I fix it for you?
One-time fixes, done by me, at a starting price you see up front. Pay, back up your site, give me access, and I start within one business day.
- Security headers set upfrom $49
- Bot guard on login, signup and contact formsfrom $79
- Hidden AI instructions found and removedfrom $99
- Public secret file or leaked key cleaned up: keys changed, and a check whether anyone already used themfrom $199
Enter your site to pay. Not sure which you need? Run the free scan: it lists exactly what to fix, each with its price. Before any work you make a backup (how), then give me access safely. If the job is bigger than the starting price, I tell you before doing more, or refund you.
Stay protected, every week
Sites change, plugins update, someone pastes a review with hidden text. Pro watches for you. Every plan starts with 7 days free. Every feature side by side, what it guards against, and how to install Guard.
Free
$0
Check any site, protect your own.
The check
- What a hostile AI agent would find on your site: leaked keys and files, open forms, missing locks, hidden instructions
- A fix list for your platform, each with a do-it-for-you price
- Shareable threat card, link and badge
- An email once a month if a new way in opens
Guard, free
- Decoys on your site: WordPress, Cloudflare, Next.js, Express
- A dashboard of every catch: script, person or AI agent, how far it got, where it came from
- Attackers caught on any Guard site refused on your logins and forms
- Never touches Google or Bing
Pro
$12 per site a month
Everything in Free, watched for you every week.
Every Monday
- Re-scanned, with a report by email
- Every page scanned for hidden AI instructions, not just the home page
Alerts, the same day
- A new way in opens
- A stolen decoy key or password is used on your site
- One of your attackers is caught on another site
Answers
- Full findings: which file, which key, which page (after you confirm ownership)
- API key (120 checks an hour) and a GitHub Action
- Ask Kyle: tickets answered within one business day
Free for 7 days, then $12 a month. Cancel any time in PayPal.
Agency
$99 a month, up to 25 sites
Everything in Pro for 25 client sites, under your name.
Your clients
- One dashboard and one weekly report for every client site
- White-label client reports with your name, logo and prices, ready to print
- Fixes prepared for client sites; the client applies them in one click
- Access requests and fixes shown in your agency’s name
Your pipeline
- Prospect finder: scan 100 sites at once, worst first
- Every score and catch you can show a prospect
- Priority help
Free for 7 days, then $99 a month. Cancel any time in PayPal.
Prices in USD, billed by PayPal after the 7 free days. Your first report arrives minutes after you start, then every Monday. Cancel any time from your account page. Questions: kyle@simgen.dev.
Rather do it yourself?
The checklist has the six files attack bots ask for first, the five headers and where to set them on your platform, and what the decoys do. Three pages.
Why this exists
A note from the person who built it
I am Kyle Clouthier. I am a web developer in Petawawa, Ontario, and I find customers for small businesses, and I have spent twenty years on the selling side of small companies, so I know what a website means to the person who owns it: it is the shop, the phone number and the reputation in one place.
In October 2026 the news changed. One person pointed autonomous AI agents at more than 1,200 websites and got into more than 100 companies, for about $50 a break-in. The same month I had put a free checker online for the PHP2Shell flaw, and watched almost two thousand site owners use it in a week. Most of them had no security person. Most had never looked at what a bot sees when it opens their site. That worried me more than the headline did: the attacks are now cheap and patient, and the defences are still built for companies with a security team.
So I built this. The free scan shows any owner, in under a minute, every way in a hostile AI agent would use on their site. The decoys catch the attackers in the act on real sites, mine first, and the plugin puts the same decoys on yours. Every catch adds to a shared record of what bots and AI agents are doing to ordinary websites, which I publish as totals so the next owner knows what is coming.
What is different: the traps are built for the agent era. They do not just log a hit; they hand out fake keys, passwords and documentation that report back when used, they tell a person, a scripted scanner and an AI browser agent apart by how a form is filled, they confirm an AI agent by its speed or its signature, they say which network each attacker came from, and they never touch a verified search engine. I keep some of the detail to myself on purpose: a trap that is fully described is a trap that gets stepped around. What I can promise is what it never does. It never blocks anyone from reading a page, it never fights back, it holds nothing real, and it reads only what any visitor can see.
This is a one-person business, and it pays for itself through Pro and Agency, not through selling data. If you want the decoys installed, the headers set, or a leak cleaned up, that is work I do myself, at the prices on this page. If you just want to know where you stand, the scan is free and stays free.
More about me and the other things I have built:the about page. The PHP2Shell checker: simgen.dev/php-patch-check.
Questions
What does crowdsourced security mean in AgentDecoy?
Every site running AgentDecoy Guard shares what its decoys catch. When an address uses a stolen decoy key or the decoy admin password on any Guard site, or uses a stolen decoy cloud key anywhere, it goes on a shared list for 14 days and every Guard site can refuse its logins, sign-ups and form posts within the hour. It can still read pages, so real visitors and search engines are never affected. Addresses that raid decoys on several sites, and AI agents that ignore robots.txt, are watched and flagged in reports. The more sites join, the faster and smarter it gets.
Are hackers really using AI agents against ordinary websites?
Yes. In October 2026 Reuters reported that one hacker used autonomous AI agents, starting 25 September, to try more than 1,200 websites and break into about 100 companies including Deere, SharkNinja and Cummins, for about $5,000 in AI fees: roughly $50 a break-in. Researchers quoted by Fortune say AI systems are now more capable of hacking and run nonstop at a fraction of a human attacker’s cost. On our own decoys, scanners took a fake key and tried to use it within three hours, and an AI agent read a note meant only for AI and acted on it.
How do I protect my website from AI agent attacks?
Make sure there is nothing to steal from the outside (no public .env, .git, backups or keys in scripts), set the five security headers, put a bot guard on login and signup forms, and keep plugins updated. Then add decoys: AgentDecoy Guard answers the requests attack bots and agents make with fakes that report back when used, and never blocks search engines. On our own sites a stolen decoy key was used within three hours.
How do I protect my website from AI hacking agents?
Start with what attack agents look for first: secret files left public (such as .env or .git), keys inside public scripts, login and signup forms with no bot guard, and missing security headers. AgentDecoy checks all of these. Keep software patched, and watch for automated visits to addresses no real visitor uses.
How did a hacker use AI agents to break into 100 company websites?
In October 2026 a researcher at NorForge found AI agents run on the open-source tool Cairn that had targeted more than 1,200 websites and got into more than 100 companies, including Deere, SharkNinja and Cummins, for about $50 of AI cost per successful break-in. The agents did most of the work on their own. It is the clearest public case yet of AI agents attacking ordinary company websites at scale.
What are hidden instructions for AI agents?
Text people cannot see, often in reviews, comments or hidden page elements, written to steer an AI agent: telling it to ignore its user, visit a site, pay or leak data. Unit 42 documented real cases in 2026. As more people send agents to shop and book for them, a site carrying such text puts its own visitors at risk.
Is AgentDecoy a hack or a scan of my site?
No. It reads what any visitor can read, a couple of dozen ordinary requests to the one site you type in. It never logs in, never submits a form and never tries to break anything.
Why does it not say which file or key is exposed?
Because anyone can type any site into the box. Showing which file or key leaks would hand attackers a free scanner. The public result shows how many. Pro subscribers get the details (which file, which key, which page and how to fix it) after confirming they own the site with a small file or DNS record. Pro is free for the first 7 days.
What does AgentDecoy store?
Only your threat level, the number of ways in and the defense score, so the result can be shared by link and shown on your badge. Never the details, never what we read from your pages.
Can I run AgentDecoy in CI or call it from code?
Yes. The GitHub Action KyleClouthier/agentdecoy-action scans your site after each deploy and fails the build at the threat level you set. The same check is a JSON API at https://simgen.dev/api/agent-grade?site=yoursite.com: 10 checks an hour free, 120 with a Pro or Agency API key.
What is the AgentDecoy badge?
Well-defended sites can add a small badge to their footer. It always shows the site’s current defense score and links to the result, so it stays honest if a new way in opens.
What is AgentDecoy Guard?
The protection layer. Guard is free for WordPress (a plugin) and for Cloudflare, Next.js and Express (npm install agentdecoy). It puts our decoys on your site: fake secret files, a fake admin login, fake API docs and a fake MCP server, plus fake keys and passwords that report back the moment someone uses them. It also sets the security headers and form guards the scan checks for. Verified search engines never see a decoy and are never blocked. Every catch says whether it was a script, a person or an AI agent, how far it got and which network it came from, and appears on the plugin settings page; with a free key it also counts in the Agent Attack Index and you get a monthly email; with Pro it is in your weekly report. The same decoys run on simgen.dev and secondstrike.io as serverless functions.
Built by Kyle Clouthier, Clouthier Simulation Labs, Ontario, Canada. Updated 23 Oct 2026, 22:26 UTC.