Class details

You’re registered

AI Security & Infrastructure — individuals, teams & regulated data

AI Practicum · hosted by JP James

How to join

Join Microsoft TeamsAdd to calendar (.ics)
When
Thursday, October 15, 2026
Time
9:00–10:30 AM ET / 6:00–7:30 AM PT
Where
Microsoft Teams
Meeting ID
285 762 575 793 34
Passcode
SK2yU7k2

Class library · Past classes

What we’ll cover

9:00–9:30 AM ET

Level 1 — Personal: secure ChatGPT and Claude, and keep your work local

  • How ChatGPT and Claude handle your data: business vs. personal plans, training settings, chat history, connectors and two-factor sign-in.
  • Two kinds of local: local files (Claude Cowork or OpenAI Codex work on folders on your laptop, while the model still runs in the provider’s cloud) and a local model (Ollama or LM Studio, where nothing leaves the laptop).
  • Prompt injection: when an email, document or web page tells your agent what to do, and the rules that stop it.
  • A personal security prompt every agent you use should follow, in a local-files and a local-model version.

9:30–10:00 AM ET

Level 2 — Teams: rules for your agents, tools and people

  • One gatekeeper agent for every password release and security action, with a one-time code (OTP) or two-factor approval; financials, outbound messages and confidential files also need a master passphrase.
  • Payment fraud and deepfakes: invoices, bank-detail changes and payments are approved through a channel separate from email, with a call-back to a known number, never on an email, call or video alone.
  • A security-scan skill that inventories email, files, messages and agent access into one encrypted, two-factor location, and expiring secure links instead of attachments.
  • Offboarding and a kill switch: remove a person’s and an agent’s access in one step, and stop every agent at once when something goes wrong.
  • Staff using personal AI accounts: an AI acceptable-use policy you can adapt, and the prompts to propagate across every agent.

10:00–10:30 AM ET

Level 3 — Regulated data: enclosed infrastructure and vendor terms

  • Running models inside Amazon Bedrock, Microsoft Azure AI Foundry or Google Cloud so the whole ecosystem stays enclosed in your own cloud account.
  • Gating models and systems to closed virtual machines reachable only from specific laptops, IP addresses and physical locations, still behind two-factor, with walled, controlled outbound file transfer.
  • Vendor terms to check: training on your data, retention, zero data retention, data processing agreements, and business associate agreements (BAAs) for HIPAA.
  • When you need it (PII, account and routing numbers, HIPAA data), and when a regulated third-party vendor is the better answer.

Open the class guide: checklists and agent security prompts