Learning path
AI security
Prompt injection, AI agents with tools and MCP servers, data leaking through assistants, the AI supply chain, and how a SOC monitors all of it. Built on the OWASP Top 10 for LLM Applications 2025, MITRE ATLAS and ATT&CK v20.
Start this path Free, in the browser, no account needed.
The steps
- Theory and check · 9 minPrompt injection, direct and indirectTell direct from indirect prompt injection, and see why the indirect kind reaches the SOC.
- Theory and check · 10 minAgents, tools and MCPExplain how an AI agent turns an injection into an action on a real system.
- Theory and check · 9 minThe OWASP Top 10 for LLM Applications 2025Name the ten LLM risks and recognise each in an incident description.
- Guided case · Advanced · 35 minThe coding assistant read an SSH key
- Hands-on lab · simulated machineAI agent abuse: the helpful tool
- Theory and check · 8 minData leakage through AI assistantsSpot the ways company data leaves through AI tools, and the controls that see it.
- Theory and check · 8 minThe AI supply chainTreat models, datasets and MCP servers as software you install, with the same risks.
- Theory and check · 10 minMonitoring AI agents in the SOCWrite a first hunt for an AI agent doing something it should not.
- Theory and check · 7 minMITRE ATLAS for AI attacksMap an AI attack to ATLAS, and know when ATT&CK covers it too.
- Open case · Advanced · 35 minnpm install, and then the cloud keys moved
- Hands-on lab · simulated machineSupply chain on a developer box
Part of a bigger plan
Every theory unit here is also part of the SOC Analyst level 1 roadmap, with spaced reviews and capstones.
Other paths
This year’s attacksThe patterns 2026 incident response saw most: help desk calls, ClickFix, fake updates with remote access, npm worms, exploited VPN appliances, fake IT workers, hypervisor takeovers, hijacked AI coding agents, fake help desks on Teams and workflows that steal CI secrets. Every attempt is a different version.
Identity and emailSign-in logs, MFA, stolen session tokens, inbox rules and app consent: how accounts get taken over, and how you prove it.
Windows and Sysmon loggingEvent IDs, logon types, process trees and scheduled tasks: read an endpoint the way an analyst does.
Network, proxy and DNSVPN, proxy and firewall logs: spot the outside getting in, and the inside reaching out.
Incident responseScope, contain and report: what you do after the alert turns out to be real.
Windows endpoint, hands-onA simulated Windows box in your browser: run real commands, open Event Viewer and Task Manager, and investigate like an analyst. First look, process trees, event logs, PowerShell, ClickFix and persistence.
Email and phishing analysisRead a header like a mail server does, know what SPF, DKIM and DMARC prove, take a lure apart safely, and recognise the phishing that passes MFA.
Cloud and SaaS identitySign-in records, stolen sessions, OAuth apps with too much access, help desk takeovers and cloud audit logs: the attacks that need no malware at all.
Linux monitoringauth.log and the journal, cron and systemd persistence, authorized_keys, reverse shells and cryptominers: the servers behind most of the internet.
Threat intelligenceFrom indicator to behaviour: enrich hashes, IPs and domains honestly, share under TLP, and turn a vendor report into sweeps and hunts mapped to ATT&CK.
Detection engineeringWrite the rules the next analyst gets alerts from: query in KQL and SPL, read and write Sigma, tune without blind spots, and prove coverage with tests.
Ransomware responseCatch it before encryption: the precursors, the first hour, evidence before eradication, and who has to be told by when.