UnHack with Drex DeFord

November 9, 2026: Bill Russell, Drex DeFord, and Lori Boisjoli, in Lori's first Newsday as a 229Project principal, back from regional staff roundtables in Boston and Charlottesville.
An AI agent that logs in with someone's password is that person, as far as an auditor can tell. Drex DeFord expects the question that follows: why did you open this patient's record 250 times? "That wasn't me, that was my bot" won't be accepted. Bill Russell runs 12 GrokBots at 229Project, and his first problem was whether that is one identity or 12.
Lori Boisjoli would treat each bot like a new hire: a role, access to specific systems, an audit history, and a way to terminate it. The same agents are already working in physician practices and in vendors connected to health systems, often on a person's credentials. Bill gives it about three months before they are everywhere.
Key Points:
0:00 Auditors won't accept "that was my bot"

0:28 Roundtable takeaways and a bot for governance

4:51 EHC Consulting

5:18 October's outages and the holes AI keeps finding

8:28 Island

8:56 Know what happens when an app goes offline

12:41 Rackspace

13:29 Who pays and who's liable for AI in care

23:22 One bot or 12? Registering AI agents

30:43 Bots that talk to each other, and bots you don't control

33:34 See you next week

LinkedIn: 229Project
Donate: Alex's Lemonade Stand: Foundation for Childhood Cancer

On Monday, October 26th, the University of New Mexico Health Sciences Center spotted a cyber event in Albuquerque. By about 12:30, ambulances were being turned away from UNM Hospital — New Mexico's only Level I trauma center, burn unit, and children's hospital — and for about two hours the most seriously injured people in the state had to go somewhere else. That evening a new ransomware gang calling itself Byod claimed the attack, saying they took out 36,000 systems, Active Directory, the entire Microsoft Entra tenant, and walked away with Social Security numbers, payroll, patient information, and Cerner EHR rosters. UNM says the attack hit some end-user devices but did not penetrate core systems or servers, critical systems were coming back online by the 27th, and it made no concessions. Then Byod posted again — this time to apologize, offer free decryption keys for hospitals only, and say they would pay affected patients up to $2,500 each. Tucked into that apology was a line that said they were still going to keep the data, and the offer asked patients to provide proof over Session, an encrypted messaging app. Drex's read: that looks a lot like a phishing lure. UNM moved fast, warning patients and staff not to respond and to report contact to the FBI at ic3.gov. Health systems plan for downtime and diversion; what they often don't plan for is the week after, when attackers or copycats contact patients with fake refunds and fake billing calls. If you get hit tomorrow, is your patient scam warning already written and ready to post on day one?
Thanks to Zscaler Healthcare for sponsoring today's show. Drop Drex a note at drex@229project.com.
Remember, Stay a Little Paranoid.

November 2, 2026: Bill Russell, Sarah Richardson, and Drex DeFord, back from regional roundtables in Kansas City and Los Angeles and an event in San Diego.
Most health systems are still at level two of Bill Russell's five levels of AI use: asking AI questions and handing it tasks through Copilot. Level three gives an agent a job that can run through the night. Level four is an AI workforce, and his own company is moving there now. He runs about a dozen bots, among them three for development and one that works as a CFO.
That workforce is arriving faster than the controls around it. Drex DeFord tells users not to hand their passwords to an agent and wants IT to make the right way the easy way. Epic now allows agents to use a computer, as Bill heard it recently, and Drex expects Epic's caution to show up in what an agent may write into the record. The harder health systems push into agents in 2027, the more the brakes matter.
References:
The GenAI Juggernaut: U.S. Healthcare Is Not Prepared (January 2025)

Healthcare's Oppenheimer Moment

Key Points:
0:00 Healthcare is only at AI level two

0:19 Roundtables and AI resiliency

4:14 Tegria

4:37 Take, shape, make, and Copilot

6:55 Censinet

7:29 Five levels of AI use

13:08 Rackspace

13:55 Bots, passwords, and going offline

21:35 The acquisition boom

27:39 Epic agents and the brakes

29:53 Who monitors AI after go-live

38:28 The Larsen papers

41:16 See you next week

LinkedIn: 229Project
Donate: Alex's Lemonade Stand: Foundation for Childhood Cancer

PHP is software most people never think about, yet it runs roughly 70% of the websites we visit, including a lot of hospital sites, portals, and login pages. Last week, details of two critical PHP flaws leaked before the fix was ready, PHP rushed patches out a day early, and the NSA and Europe's cyber response team told organizations to patch right away or take systems offline. The twist: the bug had been spotted before but looked too hard to weaponize, until researchers pointed AI at it and reportedly built a working exploit in about a week. Drex also covers a poisoned LiteLLM release, a compromised certificate authority, and a Node.js flaw, all in the same 48 hours. How many systems in your organization run PHP, who owns each one, how many are on versions that will never get another fix, and when the next bad day hits, can you patch in hours, or does it take a committee?
Thanks to SureTest for sponsoring today's show. Drop Drex a note at drex@229project.com.
Remember, Stay a Little Paranoid.

October 26, 2026: Bill Russell and Sarah Richardson, back from the CTO, CIO, and CISO summits at Lake Oconee.
At Lake Oconee, Bill Russell heard a lot of AI governance talk. Some of it sounded responsible. Some of it sounded like a shield held up by leaders who find AI overwhelming and want the pace slowed to one they are comfortable with. In his view, AI governance should speed up the adoption of technology.
His advice to CIOs is to start in IT and leave clinical for later. Put the network configs and security settings in one repository an AI can read, and in about 15 minutes it can list the holes and the certificates that have expired, before a server goes down. Sarah Richardson named the expired certificate tied to the service account of someone who no longer works there. Health systems have never trusted people with the medical record without controls. Bots get the same job.
Register for UST's Oct 29 webinar with J.D. Whitlock and Jason Buck: https://events.ust.com/Epic-on-Azure-IRE
Key Points:
0:00 Governance should speed adoption up

0:29 Lake Oconee: governance or a shield

4:41 Clearsense

6:30 AI finds what sits in plain sight

9:12 Censinet

9:52 300 projects and the right yes

13:57 UST

14:37 Rush, Elisity, Epic, UMass

22:37 Security, tech, and the bench

29:29 My admin is a bot

32:46 You're just flat-out wrong

35:36 See you next week

LinkedIn: 229Project
Donate: Alex's Lemonade Stand: Foundation for Childhood Cancer