Skip to content
Hive Research Institute

Economics · Artificial Intelligence · Frontier Research

Learn · AI Practicum

AI Practicum · Monthly class · Thursday, October 15, 2026

AI security for teams that already use AI

What happened in the October class, in order, with the recording. Each topic that teaches something you can put to work has its own module, with the video for that part and a step-by-step game plan.

Thursday, October 15, 2026 · 9:00–10:30 AM ET · Microsoft Teams · Taught by JP James, with guest Rani Haddad of Ghidorah

Recording

Watch the class

Video coming soon. This part of the class is being edited.
Full class recording · 1:31:00

Learn each topic

7 modules from this class

Each module stands on its own, with the video for that part of the class, a game plan and prompts to copy.

0:00 in class · 10 minLet AI write your prompts, then give it limitsHave AI draft the prompt from material you already have, then set the outcome and limits so it knows when to stop.Open module →0:03 in class · 20 minSet up approved AI accounts and an AI task forceGive people business AI accounts, a monthly task force and a handbook section, so AI use is visible and safe and aimed at real returns.Open module →0:08 in class · 15 minRun AI on regulated data inside your own cloudKeep personal, financial and health data inside a cloud account you control, choose between AWS, Azure and Google, and wall off agents.Open module →0:18 in class · 30 minBuild a Risk Sweep skill that finds loose secretsTurn a checklist you already follow into a skill that scans your files, email and chats every week for passwords, keys and account numbers.Open module →0:21 in class · 15 minManage AI connectors in one placeSee what every AI tool can reach, remove what nobody uses, and run the rest through one connection layer with an owner for each.Open module →0:38 in class · 25 minDeploy a narrow agent: shadow mode and guardrailsGive one agent one job end to end, inside the tools your team already uses, and prove it in shadow mode before it goes live.Open module →1:20 in class · 25 minPut every message, password and payment behind your approvalRoute sending, logins and payments through single bots that need your approval, and paste one set of rules into every agent.Open module →

What happened

The class, in order

0:00

Opening from Pittsburgh. JP shows how AI wrote this class's deck prompt from his invitation email, and why it then produced 64 slides. Module →

0:03

Pay for business AI accounts, run a monthly AI task force and update the employee handbook. Module →

0:08

AWS, Azure or Google for regulated data. Lindon Hayes and Igor Tomic describe their AWS setup with the Kortix agent orchestrator. Module →

0:18

What a skill is, and JP's weekly Risk Sweep skill. Homework: scan your files for loose passwords and keys. Module →

0:21

Composio vs Zapier: one pipe for every connector, managed per employee. Module →

0:31

Where skills should live, and why you still need a security expert (question from Trevor Hardy). Module →

0:38

Guest Rani Haddad, Ghidorah: narrow agents, shadow mode, two case studies and guardrails. Module →

1:04

Q&A: measuring AI impact, how long agent projects take, and standardizing tools across a 100-person company.

1:10

JP on choosing tools, staffing the task force and aiming for ROI over efficiency. Module →

1:20

Running WhatsApp through GrokBot in draft mode (Lindon Hayes). Module →

1:24

JP's security demo: one Messenger bot, one security bot, codes for payments, and scans of meeting recordings. Module →

From the room

Questions from the class

How is this different from Composio?

Composio is one pipe for every app your agents touch, managed per app and per employee. JP uses it for everything; the first 100,000 actions a month are free.

0:14 · answered at 0:21
Where should my skills live if I run everything in GrokBot?

In their own folder on OneDrive, Google Drive or GitHub, with every tool pointed at it. Sweep GrokBot's messages and stored files daily.

0:31
How do you measure AI impact outside of code?

Count what the agent does: reports, ERP actions, emails, customer conversations. Then measure the return, not only the time saved.

1:04
How long does an agent project take?

Two to six weeks, depending on how mature the company's systems are. A capable IT person shortens it.

1:06 · Rani Haddad
Our leaders use ChatGPT, Claude and GrokBot. Do we standardize?

JP doesn't consolidate; the tools change too fast. Put strong people over security and your hungriest users on the task force.

1:07 · answered at 1:10

Four AI notetakers joined this class, each through an attendee's own account. If your meetings are recorded, scan the recordings the way you scan files, and say in your policy which notetakers are allowed.

Go deeper

Guides and the next class