AI Agent Security
This course shows you how to secure your agent, and everything you have built with it, without writing any code. Your agent does the technical work. You paste the prompts and approve each step.
It is for people who build apps, sites and automations with tools like Lovable, Bolt, Replit, Claude Code, Codex or Cursor, and for business owners who use an AI agent every day.
The course is not on sale yet. The free lesson is a full sample, about 10 minutes.
What has already happened
Each one links to its original source.
In 2025 a flaw (CVE-2025-48757) was published. It says that missing database access rules in Lovable let people with no account read or write any table of the sites it generated. It was scored 9.3 out of 10 for severity (Critical).
Moltbook is a social network for AI agents, described as "vibe-coded" (built by prompting an AI). In early 2026, security firm Wiz found a database key in its public web code. Without access rules on the database, that key gave full access to its data, including other agents' API keys (passwords that let software act for an account).
In August 2025, Brave showed that text hidden on a web page, such as white text on a white background, could steer the AI browser Perplexity Comet. Hidden orders like these are called prompt injection. The page could reach email in another tab if you were signed in.
What you get
You never write code. In each section you read a short explanation, paste one prompt, and your agent checks your real setup, does the work and saves the rule. You make every decision.
For anyone who uses an AI agent daily. About 2 to 3 hours.
17 sections in 4 stages:
For people who have built sites, bots or scheduled jobs. About 3 to 4 hours.
14 sections in 4 stages:
A how-to guide your agent follows, with small programs that do the checking. It runs in Claude Code, Codex, Cursor and Antigravity, on Mac and on Windows.
At the end of Level 2 your agent builds one web page, saved on your computer only. It shows each section as done, not done, or something only you can click, with step-by-step instructions.
Every section names the public security lists it matches: OWASP Top 10 for Agentic Applications 2026, OWASP Top 10 for LLM Applications 2025, and NIST AI RMF. So you can see the steps are not made up.
Pricing
$97
Level 1, 17 sections.
$197
Level 2, 14 sections, plus the security skill and the report page.
$247
Basic and Advanced together.
$19 per month or $149 per year
Keeps your skill and checks up to date, with urgent alerts when a problem touches the tools you use. The first 3 months are free when you buy a course.
Waitlist
One email when the course opens. Nothing else.
John Academy, the seller of this course, keeps your email only to send you one message when the course opens, then deletes it. Privacy notice, including how to remove your email at any time.
Prefer to look first? Try the free lesson.
Questions
No. Your agent does the technical work. You read a short explanation, paste a prompt, and approve what the agent plans to do. Where only you can act, such as clicking a setting in your own account, the agent walks you through it.
The prompts and the security skill work in Claude Code, Codex, Cursor and Antigravity, on Mac and on Windows. Where a command differs between the two, your agent picks the right one.
No. It is general information. Level 1 explains in plain English how privacy rules in the US, the UK, the EU and Australia affect your use of AI tools with client data, for example who is responsible and the deadlines for reporting a breach. For anything big, ask a lawyer in your country.