AI Patches Gaps
We’ll start with federal confirmation. The Hacker News reports that CISA has added two vulnerabilities to its Known Exploited Vulnerabilities catalog — CVE-2024-1708 in ConnectWise ScreenConnect and CVE-2026-32202 in Microsoft Windows Shell — specifically because there’s evidence of active exploitation. They note the ScreenConnect issue carries a CVSS score of 8.4, and that federal agencies are required to apply mitigations by May 12, 2026. That’s a concrete signal: exploited flaws are not just being tracked, they’re being turned into time-bound remediation requirements.
Now layer in what’s happening on the ground. TechCrunch reports that hackers have exploited unpatched Windows vulnerabilities to breach at least one organization, according to Huntress. The piece names three Windows Defender-related flaws — BlueHammer, UnDefend, and RedSun — and notes that while Microsoft has patched BlueHammer, the other two were still unpatched at the time of reporting. TechCrunch also highlights that exploit code was published publicly by a researcher using the name “Chaotic Eclipse,” and that real-world attackers moved quickly once that code was available.
And the entry point isn’t always malware-first anymore. TechRepublic reports that attackers are increasingly using Microsoft Teams to impersonate IT help desk staff, talking employees into granting access and launching remote sessions. Microsoft observed a nine-stage playbook that starts with a simple request and escalates through reconnaissance, payload delivery via DLL sideloading, persistence through registry changes, command-and-control over HTTPS, and then lateral movement and targeted data exfiltration. The key detail here is that it’s happening inside a trusted collaboration channel, using legitimate remote-assist tooling.
Finally, note what the platform vendors are doing in response. iTnews reports Microsoft plans to embed Anthropic’s Claude Mythos Preview into its Security Development Lifecycle, explicitly to identify vulnerabilities and develop fixes faster earlier in the cycle. Microsoft says it evaluated Mythos using its own open-source benchmark for real-world detection engineering tasks and saw “substantial improvements” versus prior models.
Active exploitation is being formally tracked and deadline-driven, real intrusions are occurring in the unpatched window, social engineering is moving into trusted business chat, and major vendors are integrating AI into secure development because the pace is forcing it.
Discovery Isn’t Enough
What’s driving this shift is not that defenders suddenly got lazy, or that attackers suddenly got smart. It’s that the pace and volume of “security-relevant output” has crossed a threshold where the bottleneck isn’t detection anymore — it’s coordination. The hard part is turning a flood of findings, warnings, and edge cases into a clean, owned, repeatable operational motion that reliably ends in “fixed, verified, and documented.”
The Verge frames the new reality bluntly in its look at AI-driven vulnerability discovery — tools that can find bugs and help generate exploit paths at a speed that used to require deep expertise and a lot of time. The story walks through how automated systems are finding more issues than teams can realistically patch, and how that collapse in time-to-exploit forces work to happen faster than traditional cycles were designed to handle.
And then The Hacker News puts a finer point on the problem: Mythos didn’t just increase discovery, it widened the discovery-to-remediation gap. The article describes findings that stall in spreadsheets, tickets, or PDFs with unclear ownership, and it argues that without centralized workflow, prioritization, and closed-loop verification, “more discovery” just means “more unresolved critical exposure.” Even in the best-case scenario — even if the AI is mostly right — the system still needs a place to put the output, a way to triage it against real risk, and a way to prove that the fix actually shipped.
And governance is tightening at the same time. FedScoop reports NIST issuing final updates to guidance for protecting controlled unclassified information, emphasizing clarity, unambiguous requirements, and assessable procedures. The direction of travel is toward security expectations that are meant to be implemented consistently and evaluated.
Reprice or Absorb
For MSPs, the consequence is that security is becoming a verification-and-coordination service, but many are still packaging it as vague best effort, which means accountability and labor expand faster than scope, pricing, and contracts.
InformationWeek describes this as an “invisible labor” crisis inside IT: as AI spreads across the stack, it splinters work into new, poorly defined responsibilities — prompt management, orchestration, evaluation, governance, monitoring — and none of it maps neatly to existing roles or reporting lines. The point isn’t the org chart drama. The point is that critical operational work accumulates anyway, gets absorbed by already-stretched teams, and the organization can’t reliably see, measure, or staff it. In practice, that means the customer doesn’t have a clean internal owner for “make this secure, make this auditable, make this repeatable.” They just have a growing pile of small failures, edge cases, and exceptions that keep showing up as tickets.
Now add capacity constraints. The Reveal 2026 IT Talent Survey says eight in ten tech leaders report talent shortages are already impacting operations, and it calls out the hardest-to-fill roles as AI engineers and cybersecurity engineers right at the top. That matters because it means the customer can’t simply hire their way out of the invisible-work problem, even if they admit it exists. The market doesn’t have enough of the people who can do this work well, and even when you find them, they’re not cheap, and they don’t magically integrate across teams.
The choice. Either the MSP becomes the provider that runs and verifies the security operating layer — defines the operating rules, documents the workflows, sets the guardrails, monitors the system, and prices that as a managed service — or the MSP gets trapped being the sponge.
Why Do We Care?
The real shift is that security is becoming a continuously verified service operation, and ownership ambiguity determines who ends up responsible for delivering it. When attackers use trusted tools and AI produces more findings than organizations can operationalize, the burden shifts from detection to execution, verification, and proof. For MSPs, the exposure point is that ownership is often unclear, so contractual accountability lands before contractual language catches up. The bad decision is continuing to sell security as a vague best-effort service while quietly absorbing remediation coordination, verification, and evidence production work that was never explicitly scoped or priced. In that environment, contracts with vague terms like ‘best effort’ break down under deadline pressure, especially if patch SLAs, verification responsibilities, and exception handling are not explicitly assigned.
The MSP that wins this environment is the one that stops being the sponge and starts being the operating system. Defined scope. Documented workflows. Priced governance. Closed-loop verification. That’s a business model. The alternative is absorbing the liability of an increasingly complex environment at a price point designed for a simpler one.
What to Consider
- Reclassify Microsoft Teams as a social engineering vector in your security awareness program. The nine-stage playbook documented by Microsoft is operational, not theoretical. Update client training to include explicit scenarios where IT impersonation occurs inside Teams, and establish out-of-band verification protocols for any remote session request initiated through chat.
- Convert invisible labor into explicit contract scope. Conduct a scope audit against your current client agreements. Identify every governance, orchestration, and compliance task your team is absorbing without billing. Price it, document it, and present it as a managed security operations tier — or stop doing it and make the gap visible to the client.
- Use NIST CUI guidance updates as a sales trigger. NIST’s emphasis on unambiguous, assessable security requirements is a direct opening to reframe security conversations with clients in regulated verticals. Position your service as the operational layer that makes those requirements implementable — not just a vendor relationship.
If this trend continues, MSP contracts will standardize around “evidence obligations”—time-bound patch remediation plus continuous verification artifacts (change logs, remote-assist governance records, control attestations)—and customers will treat refusal to provide that evidence as a disqualifying security risk.