Tau
AI for IT & security · 11 lessons

AI for cybersecurity and IT support teams

Use AI to read logs and alerts, sort tickets, check scripts before they run and build incident timelines, without creating new security risks. For IT support and security teams.

Who it’s for: IT support, sysadmins, security analysts and SOC teams.

IT and security teams handle logs, alerts, tickets and scripts all day. AI can read and draft that material far faster than a person. It can also leak sensitive data or suggest a command that wipes a server, so this course teaches speed and safety together.

You'll start by sanitising logs and tickets, stripping usernames, IP addresses, hostnames, tokens and customer data before anything reaches an unapproved tool. You'll get a plain-English read of a log or alert, then confirm each claim against the raw data. You'll triage a ticket queue by impact and urgency. You'll review an AI-written PowerShell or Bash script for destructive or invented commands before a lab test. You'll also analyse a reported phishing email without opening the live payload. Phishing emails are scam messages built to steal logins.

Later lessons go deeper. You'll build an incident timeline in which every entry cites its source. You'll turn threat reports into decisions about what to patch or hunt for, and draft detection rules, then test them. You'll find shadow AI, the tools staff use without approval, and vet AI agents and connectors before allowing them. Finally, you'll red-team your own AI assistant, which means attacking it on purpose to find its weaknesses.

By the end you’ll be able to…

  • Sanitise logs and tickets before any of it reaches an AI tool
  • Review an AI-written script for destructive commands before it runs
  • Build an incident timeline where every entry cites its source
  • Test an internal AI assistant for prompt injection and data leaks

Short on time? Start with the essentials

These 4 lessons cover the most useful ideas in this course. Each one takes about a minute.

Essential 1

Sanitise logs and tickets before AI sees them

Decide which AI tool is approved for which security data, and strip usernames, IPs, hostnames, tokens and customer data from anything that goes to a tool that isn't.

Essential 2

Get a plain-English read of a log or alert

Give AI a sanitised log excerpt or alert with its context, ask what happened, what is normal and what to check next, then confirm each claim against the raw data.

Essential 3

Review an AI-written script before it touches prod

Ask for a PowerShell or Bash script with dry-run, logging and a scoped target list, then check it for destructive commands, invented cmdlets and hard-coded secrets before a lab test.

Essential 4

Build an incident timeline from alerts and chat

Merge alert exports, log extracts and the incident channel into one UTC timeline where every entry cites its source, with gaps and assumptions marked for the review.

Every lesson in this course

This course has 11 video lessons of about a minute each, from the basics to expert habits. Each one ends with a question, three takeaways and its sources.

Foundations 2 lessons

  1. Sanitise logs and tickets before AI sees themEssential

    Decide which AI tool is approved for which security data, and strip usernames, IPs, hostnames, tokens and customer data from anything that goes to a tool that isn't.

    “Logs hold usernames, IPs and live tokens. Paste them in the wrong chatbot: new incident.”

  2. Get a plain-English read of a log or alertEssential

    Give AI a sanitised log excerpt or alert with its context, ask what happened, what is normal and what to check next, then confirm each claim against the raw data.

    “300 failed logons (Event 4625) in 5 minutes: brute force or a stale service account?”

Building skills 4 lessons

  1. Triage a ticket queue by impact and urgency

    Have AI classify new tickets against your own category list and priority matrix, group duplicates of the same outage, and draft first replies, then spot-check the routing.

    “Monday, 140 new tickets. 30 are the same VPN outage. Find them in a minute.”

  2. Review an AI-written script before it touches prodEssential

    Ask for a PowerShell or Bash script with dry-run, logging and a scoped target list, then check it for destructive commands, invented cmdlets and hard-coded secrets before a lab test.

    “One unscoped Remove-Item in an AI script can empty a file share. Read it first.”

  3. Analyse a reported phishing email safely

    Extract headers, sending path, links and attachment hashes from a reported email and have AI explain the SPF, DKIM and DMARC results and the lure, without opening or uploading the live payload.

    “SPF pass, DKIM pass, still phishing. The sender owns the lookalike domain.”

  4. Build an incident timeline from alerts and chatEssential

    Merge alert exports, log extracts and the incident channel into one UTC timeline where every entry cites its source, with gaps and assumptions marked for the review.

    “Three tools, two time zones, one war-room chat → one UTC timeline, every line sourced.”

Advanced 4 lessons

  1. Turn threat-intel reports into actions for your estate

    Have AI pull affected products, IOCs and ATT&CK techniques out of advisories, then check each against your asset inventory and CISA's KEV catalogue to decide patch, hunt or ignore.

    “40 advisories this week. Which three touch software you actually run?”

  2. Draft a detection rule with AI, then test it

    Describe a behaviour in plain English, have AI draft a Sigma or KQL rule mapped to an ATT&CK technique, and test it against known-bad and normal logs to tune false positives.

    “AI writes a Sigma rule in seconds. Untested, it either never fires or never stops.”

  3. Find and govern shadow AI in your organisation

    Use proxy, SaaS-discovery and DLP data to see which AI tools staff actually use, then sanction, block or replace each one and add DLP rules for sensitive data in prompts.

    “Your proxy logs already show which AI tools staff use. Most aren't on the approved list.”

  4. Vet AI agents and connectors before you allow them

    Check an AI agent or MCP connector for the lethal trifecta (private data, untrusted content, a way to send data out), its source and its permissions, and remove a leg before approving it.

    “A fake Postmark MCP server quietly BCC'd every email it sent to an attacker.”

Expert 1 lessons

  1. Red-team your internal AI assistant

    Build a set of prompt-injection, jailbreak, data-exfiltration and over-permission tests, run them against your own assistant, and log each finding with severity, evidence and a fix.

    “Hide 'ignore your instructions' in a shared PDF. Does your HR bot obey it?”

Apps we recommend

These are the tools this course uses, and what each one is best for. We have no partnership with any of them.

Microsoft Security Copilot

Best for: Incident summaries and KQL help inside Microsoft security

Summarises incidents, explains scripts and drafts KQL hunting queries inside Defender, Sentinel, Intune and Entra, using your tenant's data.

Paid

Jira Service Management

Best for: Ticket triage and AI service agents for IT support

IT service desk with AI that groups similar requests, suggests categories and drafts replies from your knowledge base.

Freemium

PyRIT

Best for: Automated red-teaming of generative AI systems

Microsoft AI Red Team's open-source toolkit for sending attack prompts, including prompt injection, at an AI app and scoring the replies.

Free
Hands-on project

Produce an AI-assisted post-incident pack

Take a closed incident or major outage from your team, sanitised or fictional. Using approved tools, build a sourced UTC timeline, explain the key alerts, draft and test one detection or monitoring rule that would have caught it sooner, and write the post-incident summary for leadership.

What you’ll make: A post-incident pack: sourced timeline, tested detection or monitoring rule with results, and a one-page leadership summary.

Check your own work

  • Every timeline entry is in UTC and cites its alert, log line or message
  • Nothing sensitive went into an AI tool not approved for that data
  • The new rule was run against real logs and its hits and false positives are recorded
  • The summary labels facts, assumptions and open questions separately
  • Each action item has an owner and a due date

Further reading and viewing

We’ve chosen these articles, research papers, videos and guides from other trusted sources.

Guide · OWASP Gen AI Security Project · 30 min

OWASP Top 10 for LLM Applications

The standard list of LLM risks, from prompt injection to excessive agency and supply chain.

Guide · MITRE · 20 min

MITRE ATLAS

ATT&CK-style matrix of real attack techniques against AI systems, with case studies.

Guide · UK NCSC with CISA and international partners · 30 min

Guidelines for secure AI system development

Joint agency guidance on designing, deploying and operating AI systems securely.

Article · Simon Willison · 10 min

The lethal trifecta for AI agents

Why private data, untrusted content and a way to send data out make agents exploitable.

Course · DeepLearning.AI · 60 min

Red Teaming LLM Applications

Free short course on probing LLM apps for injection, data leaks and harmful output.

Questions about this course

How is AI used in cybersecurity?

Security teams use AI to explain alerts, summarise incidents, draft detection rules and pull warning signs out of threat reports. Tools such as Microsoft Security Copilot work inside the security products some teams already use. In every case, an analyst confirms the AI's claims against the raw data. The course also covers the new risks AI brings, such as data leaks and prompt injection.

Is it safe to run a script ChatGPT wrote?

Not without reviewing it first. AI can invent commands that don't exist, hard-code secrets or include something destructive. Ask for a script with a dry-run mode, logging and a limited list of targets. Then read it line by line and test it in a lab before it goes anywhere near production. The course walks through this review step by step.

Can I paste logs into ChatGPT or another AI tool?

Only into a tool your organisation has approved for that kind of data. Before you share anything, strip usernames, IP addresses, hostnames, tokens and customer data. Then check every claim the AI makes against the raw logs, because it can misread them. The course starts with this lesson. It also covers finding shadow AI and checking threat advisories against your own systems and CISA's list of exploited flaws.

What is prompt injection, and how do I test for it?

Prompt injection is when hidden instructions in a document, web page or email trick an AI tool into doing something it shouldn't, such as leaking data. The risk is highest when a tool has private data, reads untrusted content and can send data out. The course shows you how to vet AI agents and connectors for that mix. You'll also build a set of attack tests and run them against your own assistant, with tools such as PyRIT.

How long are Tau's IT and security lessons?

Each lesson takes about a minute. It's a short video followed by one question, three takeaways and the sources. The course has 11 lessons, from sanitising logs to red-teaming an AI assistant. Your first 5 lessons in Tau are free, and Tau Pro gives you every course and lesson.

Get early access to this course

Tau opens soon. Join the waitlist and we’ll email you as soon as you can start this course.