Agent Grade · free
Can AI find your site, and can attack agents get in?
One grade in three parts: found by search and AI, protected from AI attack agents, and safe for the AI agents your visitors send. Free, in under a minute.
Free. Reads only what any visitor can see. Your grade is saved for 7 days so you can share it; nothing sensitive is ever shown publicly.
Found · 40 points
Google, Bing, ChatGPT search, Claude and Perplexity: are they allowed in, and do they get the page people get? A blanket robots.txt rule or a firewall can hide you from AI search without you knowing.
Protected · 45 points
What attack agents try first: secret files left public, keys inside public scripts, login and signup forms with no bot guard, and the basic security headers.
Agent-safe · 15 points
Hidden text that tells AI agents what to do. People send agents to shop and book for them; text planted on your pages can turn those agents against your own visitors.
Live updates · 24 Oct 2026, 03:00 UTC
AI agents and websites: what is happening
24 Oct 2026
Our study of the top 100,000 websites: 11.7% block OpenAI’s training bot but only 5.3% block ChatGPT search. 2,012 sites are open to Google but closed to ChatGPT search, and 515 of those never name it at all: a blanket rule shuts it out.
24 Oct 2026
Our decoys on simgen.dev caught their first visitors within hours: a scanner on a likely hijacked server took a fake /.git/config, and Meta’s AI crawler, from a Meta-owned network, followed a link that robots.txt forbids. That rule was only hours old, so it may have used an older copy; we are watching whether it repeats.
20 Oct 2026
A 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. Researcher Snorre Fagerland of NorForge put the AI cost at about $5,000, roughly $50 per successful break-in; break-ins took a little under 23 hours on average.
19 Oct 2026
Microsoft warned that AI agents with access to company systems can be steered by prompt injection and tainted inputs into actions they were never meant to take, and that access controls should sit outside the model.
Sep 2026
Anthropic: “A majority of the operations described in this report were enabled by AI via direct execution or orchestration,” with multi-agent setups doing reconnaissance, exploitation and data theft.
11 May 2026
Google’s threat team described a suspected state-linked actor using Strix, a multi-agent penetration testing framework, to find and confirm weaknesses automatically.
2 May 2026
Researchers caught all 7 AI browsing agents they tested from how they type, scroll and move the mouse; Cloudflare’s bot detection caught 1 of the 7.
29 Apr 2026
Imperva: automated traffic made up more than 53% of all web traffic in 2025, up from 51% the year before.
3 Mar 2026
Palo Alto Networks Unit 42 documented 12 real cases of websites hiding instructions for AI agents, trying to make them pay, delete data, leak their instructions or promote phishing sites.
Our study · Top 100,000 websites
Top sites block AI training bots about twice as often as AI search bots
We read the public robots.txt of the top 100,000 websites by the official rules. 11.7% block OpenAI's training bot; 5.3% block ChatGPT search. Of the 9,313 that block OpenAI training, 4,189 also shut out ChatGPT search, so they cannot be cited there. 2,012 sites are open to Google but closed to ChatGPT search, and 515 of those never name it at all: a blanket rule meant for every bot shuts it out. 11.1% have an llms.txt file.
Gold: AI search and assistants. Grey: AI training and classic search. Share of 79,574 sites that answered, blocking the home page in robots.txt. robots.txt shows what a site asks for, not what its firewall does. Tranco list 56Z8N, fetched 23 and 24 Oct 2026.
Live decoys
What our decoys caught
simgen.dev and secondstrike.io carry decoys no real visitor ever asks for: fake secret files, fake keys, a fake login. Search engines are verified and never see them. Everything that trips them is counted here, in totals only, updated hourly. Live since 23 October 2026.
Keep your grade up
Sites change, plugins update, someone pastes a review with hidden text. Pro watches for you.
Free
$0
- The Agent Grade check, any site
- Shareable grade card, link and badge
- Owner details on request, after we confirm you own the site
Pro
$12 per site a month
- Your site re-graded every week, report by email
- An email the day your grade drops
- Every page checked for hidden AI instructions, not just the home page
- Full findings once you confirm you own the site
Agency
$99 a month, up to 25 sites
- Everything in Pro for 25 client sites
- One weekly report covering every site
- Priority help
Prices in USD. Reserving is free and takes no payment; founding members keep this price when Pro opens. Payment will be by PayPal.
Questions
How do I check if ChatGPT can see my website?
Type your site into the Agent Grade box. It reads your robots.txt the way OpenAI’s bots do and loads your home page as ChatGPT search (OAI-SearchBot), as ChatGPT fetching a page for a user (ChatGPT-User) and as OpenAI’s training bot (GPTBot), then compares what each gets with what a person gets. It does the same for Google, Bing, Claude and Perplexity.
What is the difference between GPTBot and OAI-SearchBot?
GPTBot collects pages for training OpenAI’s models. OAI-SearchBot is what puts your pages in ChatGPT search. They are controlled separately in robots.txt, so you can block training and still be found in ChatGPT search. Our study found 4,189 of the top 100,000 sites block both, so they cannot be cited in ChatGPT search.
Does blocking AI bots hurt my Google ranking?
Blocking AI training bots such as GPTBot, ClaudeBot or Google-Extended does not change your Google Search ranking. Blocking Googlebot does. The risk is the other way round: a blanket rule meant for every bot can quietly hide you from AI search. 515 of the top 100,000 sites shut out ChatGPT search that way without ever naming it.
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. Agent Grade 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 Agent Grade 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; the owner gets the details after confirming they own the site.
Do I need an llms.txt file?
It helps AI assistants understand your site but it is optional and does not change your grade. In our study 11.1% of the top 100,000 sites have one.
What does Agent Grade store?
Only your grade and the three part scores, so the result can be shared by link and shown on your badge. Never the details, never what we read from your pages.
What is the Agent Grade badge?
Sites that score A or B can add a small badge to their footer. It always shows the site’s current grade and links to the result, so it stays honest if the grade changes.
Built by Kyle Clouthier, Clouthier Simulation Labs, Ontario, Canada. Updated 24 Oct 2026, 03:00 UTC.