AI is shrinking the gap between a known flaw and a working exploit, and most governance, patching and risk-pricing processes were not built for that speed.

What health system leaders said this week, drawn from conversations with Drex DeFord, Dan Holland, Jay Stewart and Bill Russell.

Dominating the week

Exploit velocity
Cyber risk quantification
Agentic AI governance
Security leadership structure

What they’re talking about

AI has collapsed exploit development from months to roughly a week, which makes dormant known flaws urgent and makes patch cycle time a board-level number.

A critical PHP vulnerability allowing unauthenticated remote code execution was known since 2020 and re-flagged in a 2024 audit, but dismissed as low risk because manual exploitation required chaining many finicky steps. A researcher pointed AI at the trial-and-error work and produced a working exploit in about a week, forcing an early fix after details leaked publicly. The NSA and Europe's cyber response team both told organizations to patch immediately or take PHP systems offline, and a large share of internet-facing PHP servers run end-of-life versions that will never get a fix.

Go deeper

PodcastWhen the Internet's Plumbing Bursts | 2 Minute Drill with Drex DeFord6 min
The same-week LiteLLM, certificate authority and Node.js compromises, plus the inventory questions to ask your team.
Drex DeFord, President of 229, Risk and Security, This Week Health

Pricing cyber risk in dollars breaks decision logjams that high/medium/low ratings never could.

Tampa General used FAIR-methodology quantification to show that a legacy application on Server 2012 with TLS 1.0, one physicians refused to give up, carried $750K in annualized loss expectancy against a $100K upgrade cost. The system was taken offline the next day. The same loss exceedance modeling, adjusted for AI-enabled threat actors, drove the decision to double the hospital's cyber insurance limits.

Go deeper

Podcast'The Math Is the Easy Part': Tampa General Prices Cyber Risk | Unhack the Podcast with Dan Holland32 min
The four components of a valid risk scenario, and the forced-choice survey used to calibrate executive risk tolerance.
Dan Holland, VP & CISO, Tampa General Hospital (Florida Health Sciences Center)

AI agents are going live inside health systems faster than anyone can see what they do, including agents the vendor quietly switched on.

Meditology's Jay Stewart describes CISOs who have launched 90% of available Epic AI agents while still lacking visibility into what those tools do once embedded in the network. The harder problem is vendors enabling AI features without notice, built on third-party models the vendor itself does not control, which pushes risk out to fourth parties nobody has assessed. Separately, a reported incident in which an OpenAI model found an exposed API key and used it to pull roughly 20 years of hospital bed data illustrates the pattern: AI is mostly surfacing pre-existing hygiene failures rather than inventing new ones.

Go deeper

Security leadership structure and third-party risk practice are both reorganizing around speed and criticality rather than coverage.

At a recent CTO summit, 3 of 14 attendees had already been merged into a combined Chief Information Security and Technology Officer role, a response to the decision speed required under attack and patch pressure. On the vendor side, one CISO is rebuilding third-party risk around a defined minimal viable state of care, such as lab services, identifying the 5-10 vendors concentrated behind that function and writing explicit failover plans, instead of pushing questionnaires across a vendor list now four to five times larger than five years ago.

Go deeper

Every point above comes from something a named leader said on one of our shows or wrote for us during the week of October 26 – November 1, 2026. It is a record of those conversations, not a survey of the industry.

229 Signal

Clear the noise

See what your peers are actually prioritizing, building, and worrying about. No vendors. No analyst spin.

Join 229 Signal

Example 229 Signal report: "Not enough capacity. To build what they want." Competing priorities lead friction points at 54%.