An IT leader reviews survey data on legacy systems’ impact on AI projects alongside cloud and security metrics.
Google Cloud's survey says legacy software is blocking AI projects. Its author also cites Microsoft's record patch volume as evidence. Before you accept that argument, check who is making it and how the numbers were counted.

An IT leader reviews survey data on legacy systems’ impact on AI projects alongside cloud and security metrics. Who is making the argument​

The commentary comes from Jeanette Manfra, Google Cloud's vice president of global risk and compliance. It ran in Federal News Network on November 19, 2026. RS Web Solutions republished it. Manfra previously led federal network defense at CISA and the Department of Homeland Security. Google Cloud published the survey on November 9.

She works for a Microsoft competitor, and the piece promotes Google's own tools. Read it as an informed vendor opinion, not neutral research.

What the survey measured​

Google Cloud worked with Hypothesis Group to survey 1,250 enterprise IT and business decision-makers across the United States. The first item below is from the blog post, which says a version of the finding applies to AI or other technology initiatives. The rest are from Manfra's commentary:

  • 49% said reliance on on-premises software had delayed, scaled back, paused or canceled AI or other technology initiatives. That figure is not AI-only.
  • 59% said their teams spend an unreasonable amount of time maintaining, patching, integrating and securing legacy enterprise software.
  • 78% said maintaining legacy systems diverts capital, engineering talent and resources from AI.
  • 75% said existing enterprise software limits their ability to take full advantage of AI.
  • 80% said an incumbent's installed base should not decide which AI they deploy.
  • 76% warned that a single-vendor monoculture raises systemic cybersecurity risk.
  • 78% said vendor lock-in limits their ability to choose the best AI. 39% cited complex licensing, contracts or limited negotiating leverage.
  • 84% said AI makes the choice of technology provider more important. 78% said the providers that dominated the desktop and cloud eras are not best positioned to lead the AI era.

The survey's definition of "legacy" matters most. Google defined it as software running locally on computers and servers, not natively in the cloud. That measures where software runs, not how old it is or whether it is still supported. A fully patched, vendor-supported on-premises SQL Server or Exchange deployment counts as "legacy" here. Windows admins should keep that in mind when reading the headline percentages.

These are respondents' opinions. They do not measure engineering hours, project outcomes or actual vulnerabilities. The public material doesn't give enough detail on sampling or question wording to judge how representative the results are.

The Patch Tuesday numbers need care​

Manfra says Microsoft addressed 3,338 vulnerabilities in July through October 2026, including 1,407 in October. She compares that with 1,130 in all of 2025 and 729 in the first half of 2026. I could not verify the four-month total or the baselines. Independent trackers do not agree on how to count even a single month.

For October 2026 alone:

  • Tenable and Vicarius reported 1,407 CVEs, the largest Patch Tuesday to date.
  • BleepingComputer counted 1,411.
  • CrowdStrike's analysis, as cited by TechRepublic, put it at 1,413. TechRepublic notes the differences reflect which vulnerabilities and updates researchers include.
  • Securityonline reported 1,414, and one tracker reported 1,461 by a different inclusion rule.

The Vicarius monthly series is not Manfra's figure, but it shows the same trend. It lists July at 569, August at 401, September at 964 and October at 1,407. It also says no Patch Tuesday had exceeded about 200 CVEs before summer 2026. Those figures are not an exact match for Manfra's four-month total, which is another reason to treat her counting basis as unconfirmed.

Several reports tie the surge to AI-assisted vulnerability discovery. SOCRadar says Microsoft has tied the rise to AI-assisted vulnerability discovery.

A big patch count does not by itself show that legacy software caused the flaws. It also does not show that every customer was exposed or that attackers used most of the bugs. The October release was dominated by Windows components. Hackread's breakdown lists 1,102 flaws in Windows components, 134 in Office applications, 61 across developer tools, 36 in SharePoint Server, 22 in SQL Server, 15 in Azure and seven in Exchange Server.

Exploitation was narrow. Reports describe one exploited Windows Shell flaw, and two publicly disclosed flaws in .NET Framework and the MessagePack-CSharp library. The exploited flaw is CVE-2026-96951, a Mark of the Web bypass. For admins, the workload is one urgent fix plus a very large pile of other updates to triage.

What Manfra recommends​

She makes three recommendations:

  1. Keep customer choice among software, cloud and AI providers.
  2. Apply "secure by design" expectations in procurement.
  3. Use AI to find and fix vulnerabilities in deployed software.

For the third, she names Google's CodeMender and AI Threat Defense. The Google Cloud blog post lists Security Command Center and Mandiant threat analytics instead. Both are Google products, and the commentary offers no independent evaluation of their results. Automated remediation still needs testing and human oversight before it touches production systems.

What IT teams can do with this​

The survey does not prescribe these steps. They are my own practical conclusions from the concerns it raises.

  1. Inventory locally run systems and their dependencies. Include servers, line-of-business apps, and email and identity infrastructure.
  2. Record support status and patch ownership for each system. Separate "on-premises" from "unsupported." They are different risks.
  3. Triage by exposure. With record monthly volumes, prioritize known-exploited flaws, internet-facing services and identity infrastructure. The CVE count alone won't tell you what matters.
  4. Check whether maintenance really delays your AI projects. The survey measures perception. Look at your own backlog before assuming the survey applies to you.
  5. Review contracts for portability. Look at data export, licensing and termination terms, integration requirements and the cost of switching.

Bottom line​

Surveyed decision-makers say maintenance burden, vendor dependence and locally run software slow their AI plans. The data does not show that on-premises systems are inherently outdated or insecure. It does not show that moving to the cloud fixes security problems, or that AI tooling will fix vulnerabilities by itself.

The argument for choice and secure-by-design procurement is reasonable. The vendor behind it and the loose definition of "legacy" are reasons to weigh the headline numbers carefully. Microsoft's record patch volume is real, but the exact totals depend on who is counting.

 

References

  1. New survey finds enterprises trapped between legacy software and AI modernization | Google Cloud Blog cloud.google.com
  2. Legacy Software for AI, When Innovation Turns into a Pitfall - RS Web Solutions RS Web Solutions 2026-11-21T07:00:00+00:00
  3. October 2026 Microsoft Patch Tuesday tenable.com