| Managed AI Agents
Introducing Managed AI Agents
The busywork help your team wants, under the same discipline we bring to HIPAA and SOC 2.
Your people are already curious about AI. Some are already using it. The question most owners ask us isn't "should we?" It's "what can it touch, and who's checking?"
AI agents don't just answer questions. They read inboxes, draft replies, move meetings, and connect to the tools your business runs on. Set up in a hurry, an agent can see more than it should or send something nobody checked. Most small businesses don't have an IT department to sort that out.
So we built Managed AI Agents. We deploy, onboard, and manage Grok Bot agents for small and mid-size businesses, and we run them under the same rules we use for our own.
The rules come first
One job, one owner. Every agent has a defined job and a named person accountable for it.
Agents draft, humans send. An agent can write the email or the reply. A person reviews and sends anything that leaves your business.
No PHI. Patient data doesn't go into an agent, and we scope its connections to keep it out. Nobody pastes client PHI into a chat.
Least privilege. Each agent starts read-only or with the least access its job needs.
Everything is written down and logged. Owner, purpose, access, and activity are documented and reviewed, so you have an answer when a client or auditor asks.
How it works
Talk it through. A short call about your team and the data you handle.
Readiness review. We look at your accounts and the rules that apply to you, including HIPAA and Texas HB 300, before anything gets connected.
Set up and onboard. Least-privilege access, written rules, and training for each person.
Manage and review. We check access and activity on a schedule and switch off what's no longer needed.
Why we're confident
We made ourselves the first customer. VanRein runs agents in our own firm every day. They triage email and Slack, prepare first-pass BAA reviews, and watch CISA vulnerability alerts. Each has one job and only the access it needs, and a person approves anything before it goes outside the firm.
Reach out for details on pricing and packages.
Watch it work, live. Episode 2 of our new X Live show is a live demo of Managed AI Agents: a real agent drafting, a real person approving. Watch for our X Live invite. If you missed Episode 1 on VRC Assurance, we'll share a replay.
| This Week's CVEs from CISA
Two new items since Tuesday's issue, and the first one is today's work. On November 17, CISA put out an alert urging everyone to patch a critical Java flaw immediately. Java often comes bundled inside other software, so it may not show up on your software list. Ask your IT provider and the vendors behind your business apps whether their products include it.
CVE-2026-114938 | Oracle Java SE and JDK (the HotSpot Java engine, nicknamed "Java4Shell"). Specially crafted input can crash Java and let an attacker with no login run code on the system. Patch or apply the vendor mitigation. Not yet in the KEV catalog, so no CISA due date, but CISA says patch now.
CVE-2026-115731 | Perforce Puppet Enterprise (IT configuration management). A machine with a signed agent certificate can run code on the Puppet server, and if the server signs certificates automatically, no login is needed. Patch or apply the vendor mitigation. KEV due date: December 1, 2026.
For our energy clients: CISA also posted 12 industrial control system advisories on November 17, mostly Siemens products. None are on the exploited list.
Why it matters for HIPAA: software hidden inside other software still belongs in your risk analysis, and your patch management records should show when each fix went in, including on systems your vendors look after.
| Framework Spotlight
HITRUST Now Covers AI, Too
If you sell into healthcare, HITRUST may already be on your customers' list. What many teams miss is that HITRUST now offers assessments built for AI, on the same HITRUST CSF behind e1, i1 and r2.
AI Security Assessment and Certification. Prescriptive security controls for AI systems and platforms that are already deployed. You can earn it on its own or pair it with an e1, i1 or r2.
AI Risk Management Assessment. 51 controls focused on AI governance and risk, harmonized with NIST AI RMF and ISO 23894.
Which fits? If you build or host AI that touches customer data, the security certification is the one buyers are likely to ask about. If you need to show your AI governance holds up, the risk management assessment is the place to look. And if ISO 42001 is also on the table, it's worth sorting out which one your customers actually want before you start.
On the podcast: In "Compliance, Elevated: The Next Chapter of VanRein", we talk through what clearer communication and a smoother audit flow look like for growing teams.
Before your first agent: Take the AI Exposure Score to see where AI already touches your data, and if you need the policy and guardrails in place first, start with Secure AI Adoption.
Or just reply to this email and tell us which job you'd hand an agent first.
Learn more: vanreincompliance.com/services/managed-ai-agents
Grok Bot is a trademark of its owner. VanRein Compliance is an independent compliance firm.
Thanks for reading, and for trusting us with your compliance.
Rob and Dawn Van Buskirk
Co-founders, VanRein Compliance
Compliance, Elevated.
P.S. Already using AI tools on your own? That's fine, and it's common. Reply with the list, and we'll tell you which ones belong on your vendor list and which ones need a BAA.

