The industry’s first reflex this week is to talk about agents. Microsoft announced ISOC in Defender, a foundation for agentic security that folds SIEM and threat-protection into one story, and a lot of leadership decks will follow that headline. Your cybersecurity queue should not. Nine critical defects in Adobe Connect and Adobe Experience Manager Forms landed in the same news cycle, with arbitrary code execution and privilege escalation on software that still hosts all-hands, partner webinars, and public-facing intake forms. Those boxes already have accounts, sessions, and a path into the rest of the estate.

You already know how this movie goes. A vendor ships a console with a new noun. Analysts ask whether your SOC is “ready.” Meanwhile the webinar server you stood up in 2019 is still reachable, still running a fat Java or application stack, and still treated like furniture. That is the box that will page you. Treat the agentic pitch as optional reading. Treat Connect and AEM Forms like production.

The Webinar Box Is Still a Privileged Host

Connect is the product people schedule. AEM Forms is the product people forget. Both sit in that cursed middle zone of enterprise software: important enough to be internet-facing, boring enough to miss the monthly patch meeting. SecurityWeek’s write-up is blunt about the impact class. Arbitrary code execution. Privilege escalation. Nine critical defects. You do not need a novel exploit write-up to know what that combination means on a host that accepts uploads, renders content, and authenticates outsiders.

Walk your own estate for a minute. Who owns the Connect cluster? Marketing? A training vendor? A regional office that bought it to stop paying Zoom overage fees? Ownership drift is how these hosts go stale. The last person who knew the admin password left in 2024. The SSL cert auto-renews. The hypervisor snapshot is “good enough.” Nobody put it on the crown-jewel list because it is not the domain controller and it is not the ERP. Attackers do not need it to be. They need it to execute code, then look for the service account, the file share with recordings, the SSO integration, and the management port you left on the same VLAN as everything else.

AEM Forms deserves the same suspicion. Forms products collect the data your privacy policy pretends is tightly scoped: names, case numbers, uploads, sometimes government IDs, sometimes health or finance context you never meant to park next to a public URL. A privilege-escalation bug there is not a theoretical CMS problem. It is a data-handling problem with a shell. If you run author and publish tiers, assume the internet-facing publish node is the beachhead and the author node is the prize.

Defense in depth on paper already says this. Segmentation. Least privilege. No shared local admins. In practice, collaboration appliances get a firewall exception labeled “video” or “forms” and a monitoring profile copied from a random web server. That is how you get a silent root path with a friendly calendar invite on the front.

Agentic Consoles Do Not Inventory Forgotten Software

Microsoft’s ISOC announcement will dominate internal Slack because it is designed to. SIEM and threat detection in one narrative. Agents that propose work. A SOC that looks like a product roadmap instead of a staffing problem. Fine. You can evaluate that on its own timetable. You cannot let it become the reason Connect stays on last quarter’s build.

This is a bad look for any security program that measures maturity by console count. You can buy a prettier incident queue and still have an unauthenticated-looking attack surface on a meeting platform that every employee will click without thinking. Users trust the invite. Help desk trusts the “join with a code” flow. Your threat-protection stack trusts the vendor signed package. None of that inspects whether the appliance behind the URL is patched. If your detection story starts after the host is owned, you are doing forensics, not prevention.

Stop treating “it’s just meetings” as a risk rating

Risk ratings follow embarrassment, not architecture. Domain controllers get love. VPN concentrators get love after the last breach slide. Webinar and forms servers get a shrug until a researcher publishes a screenshot of a webshell in the recordings directory. Change the rating now. If it terminates TLS for outsiders, stores credentials or session tokens, or runs as a privileged service account, it is a tier-1 identity-adjacent system. Put it on the same patch SLA as your VPN and your IdP. Cyber security programs that still score these hosts as “low, internal comms” are grading the org chart, not the packet path.

You will hear that the agentic SOC will “correlate” this class of issue. Correlation is not inventory. If the asset is missing from CMDB, missing from the vuln scanner’s credentialed list, and missing from the jump-host allowlist, no model is going to propose the right ticket. Humans still have to name the box.

Cybersecurity Hardening Starts With the Collaboration Stack

Here is the work that actually reduces risk this week. No new platform required. No waiting on a SIEM migration. Do the immediate items in the first two business days, then keep the ongoing controls on the same calendar as identity and backup tests.

1. Name every Connect and AEM Forms instance, including vendor-hosted and “temporary” event clusters. Pull DNS, certificates, reverse-proxy configs, and procurement. If a regional team can stand up a meeting site without your knowledge, you do not have an inventory. You have folklore. Record version, owner, identity integration, and whether the admin console is on the public internet.
2. Patch, then prove the build. Apply Adobe’s critical fixes on a maintenance window you would accept for a VPN concentrator. Do not take a vendor “compatibility hold” at face value without a written residual-risk owner. After the change, verify the running version from the host, not from a dashboard screenshot. If you cannot credential-scan the box, that is a finding, not a footnote.
3. Shrink the path that brute-force and opportunistic scans actually hit. Admin consoles go behind SSO, phishing-resistant MFA, and an allowlisted management network. Public join URLs should not share a server with authoring, plugin upload, or cluster management. If the product still offers local accounts, disable them or vault them with break-glass monitoring. Your edge firewall should not publish the management plane because “the meeting link needs to work.”
4. Cut the blast radius before the next advisory. Dedicated service accounts, no domain admin anywhere near the app pool, recordings and uploads on storage the web process cannot jump from, and egress that cannot freely reach your identity provider’s management API. If Connect or AEM can talk to everything behind the same internal firewall rule, you built a beachhead with a friendly UI.
5. Pre-stage incident response as host compromise, not a bad webinar. The runbook should include: take packet and disk evidence, rotate every secret the host could have touched (SSO certs, SMTP, storage keys, local admins), invalidate sessions, and hunt for webshells, unexpected scheduled tasks, and outbound beacons. Tell help desk that “people cannot join the all-hands” might be the cover story for a locked admin console. Page the IR lead, not only Facilities.

Ongoing, put these products on the same security hardening loop you already claim for internet-facing apps: monthly version diffs, certificate and plugin inventory, authenticated vuln scans, and a quarterly access review of who can publish an event or a form. Log admin actions and file writes somewhere your SOC already watches. If threat detection has no parser for the product, start with host telemetry: new processes, unexpected listening ports, and outbound connections from a box that should only talk to your IdP, storage, and a small set of CDNs.

None of this requires you to reject SOC modernization. Evaluate ISOC when you have time and a real architecture question. This week’s unpaid bill is the collaboration stack you already run. Patch it. Hide its admin plane. Assume a critical advisory will show up again, because this class of software has a long memory and a large install base.

Frequently Asked Questions

Do we have to patch Connect if we only use it for internal all-hands?

Yes. Internal users still click join links, and internal networks are where stolen session cookies and service accounts pay off. A code-execution bug on an “internal” meeting host is a lateral-movement problem with a calendar invite attached. If you cannot patch this week, take the admin plane off the network users actually reach and put a named risk owner on the delay.

Is vendor-hosted Adobe a pass on this work?

No. You still own identity integrations, which groups can create events, where recordings land, and how fast you can revoke SSO. Ask the provider for the patched build and for evidence, not a status page paragraph. If they cannot tell you the version, treat that as an incident-response input and restrict what the tenant can access in your directory.

Should we pause SOC tooling evaluation until these hosts are clean?

You should not freeze every architecture discussion. You should refuse to let a console refresh outrank a critical patch on a public or partner-facing application. Rank work by reachable code execution, not by how new the acronym is. A patched meeting platform and a boring SIEM will beat an agentic demo sitting in front of an unpatched forms server.

Sources

• Adobe Patches Critical Flaws in Connect, AEM Forms
• Reimagining the SOC for the agentic era in Microsoft Defender