A cybersecurity analyst monitors server alerts and cloud infrastructure on multiple screens in a data center.
Google Cloud's pitch is that legacy software is an "innovation tax" on AI. It is a sponsored argument, not neutral research. It also leans heavily on Microsoft's Patch Tuesday numbers, and those numbers need a closer look.

The argument comes from Jeanette Manfra, Google Cloud's vice president of global risk and compliance. She is a former CISA and DHS cyber official. Federal News Network ran her commentary on November 19, 2026. It is an op-ed from an executive at a Microsoft competitor, and it recommends choices that favor her employer. IT admins should read it as an advocacy brief with some useful data, not as an audit.

A cybersecurity analyst monitors server alerts and cloud infrastructure on multiple screens in a data center. What the survey says​

Google Cloud partnered with Hypothesis Group to survey 1,250 enterprise IT and business decision-makers across the United States. The study defined "legacy" technology as software running locally on computers and servers rather than natively in the cloud. That definition matters. It is about where software runs, not how old it is. It also doesn't cover every difficult-to-modernize system or all technical debt.

The headline numbers, as the commentary and the full report present them:

  • 59% of IT leaders say their teams spend an unreasonable amount of time maintaining, patching, integrating and securing legacy enterprise software.
  • 78% say maintaining legacy systems diverts money, engineers and resources from AI.
  • 75% say their existing enterprise software limits their ability to take full advantage of AI.
  • 80% agree an incumbent's installed base shouldn't decide which AI an organization uses.
  • 78% say vendor lock-in limits their ability to pick the best AI.
  • 76% agree that depending on one vendor raises cybersecurity risk.
  • 39% cite complex licensing, contracts or limited negotiating room.
  • 84% say AI makes the choice of technology provider more important than ever.

Other coverage of the survey adds figures from the full report. 49 percent said reliance on on-premises software had delayed, narrowed, paused or canceled AI or other technology projects. The full report also says only 48% consider their stack ready to deploy AI at scale.

These are agreement responses and self-reported perceptions. They don't measure engineering hours, budgets or incidents. Nothing in them shows that a given vendor caused a breach, or that a multi-vendor estate is automatically safer. The survey also covers private enterprises in sectors such as financial services, healthcare, retail and legal. It is not a government-only sample, even though the commentary leans on public-sector framing.

The Microsoft patch-count claim​

Manfra writes that Microsoft addressed 3,338 vulnerabilities across 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.

The October figure checks out against Tenable's Patch Tuesday count of 1,407 CVEs. Computerworld's rolling coverage reports a slightly different 1,408. Tenable also counted 69 critical flaws and two zero-days, one exploited in the wild.

The rest of the four-month total doesn't reconcile with Computerworld's monthly tallies:

Month (2026)Computerworld figure
July722 CVEs, after setting aside 427 Chromium upstream relays
August751 CVE entries
September963 CVEs
October1,408 CVEs

Those four monthly figures add up to roughly 3,840. Manfra's 3,338 is about 500 lower. Counting method probably explains much of the gap, and I haven't confirmed the cause. Counts shift depending on whether upstream Chromium relays, third-party components or republished entries are included. The practical lesson is to record which counting method any patch statistic uses before quoting it.

The trend is still real, and Computerworld's coverage conveys the strain. July was described as roughly three times a normal cycle. September was called Microsoft's biggest of 2026 to that point, with two exploited flaws. October's coverage flagged a SharePoint flaw that Microsoft had marked "not exploited" but that was reportedly under attack. July also brought an end-of-support collision, with SharePoint Server 2016/2019 and SQL Server 2016 all reaching end of support.

A high CVE count is not a clean measure of insecurity. More fixes can reflect more scrutiny, including AI-assisted discovery, and more disclosure. It doesn't map directly to how many patches your environment must deploy. It does show why on-premises Windows, Exchange, SharePoint and SQL Server shops feel the pressure. Computerworld's rolling advice for several months was "Patch Now" for Windows and Office.

Where the argument is weaker​

Manfra's case has three soft spots.

  1. The framing is commercial. Google sells cloud and security products, so "legacy" being the problem and cloud being the answer is convenient. Google's own blog recommends demanding architectural and provider choice and shifting to AI-assisted defense.
  2. Cloud isn't automatically safer. The commentary says "monoculture" raises systemic risk, which is plausible. But a single cloud provider is also a concentration of dependency. The 76% figure is agreement, not an incident rate.
  3. AI remediation is promoted, not proven. The commentary names CodeMender and AI Threat Defense as Google's tools. The research summary doesn't validate them independently. Automated fixes to production systems still need human review and approval, which the full report also stresses.

The report is also less binary than the op-ed. It treats modernization as building an environment that can adapt to new models without giving up reliability, security or governance. It lists data quality and access, integration, migration, governance, operating costs and portability as constraints alongside maintenance.

What IT teams can take from it​

You can use the survey's themes without buying the sales pitch. A practical checklist:

  • Measure the burden. Track actual hours spent on patching, integration and workarounds. A perception survey won't tell you what your own teams spend.
  • Check data readiness first. Many AI projects stall on data access, quality and governance, not on where the application runs.
  • Test portability. Ask what it would cost to move your data and workloads off any provider, including the cloud one.
  • Read licensing terms. The 39% who cite complex contracts suggests it's worth checking AI add-on bundling and renewal terms.
  • Make secure-by-design a procurement criterion. Ask vendors about default configurations, patch cadence and end-of-support dates before you sign.
  • Prioritize patches by exposure. Internet-facing and exploited components such as SharePoint and Exchange come first. Don't chase raw CVE totals.

The bottom line​

The commentary has a real point: patch workloads and lock-in are real costs that compete with AI projects for the same people and budget. Microsoft's 2026 patch volumes give that point weight, and the October count holds up against independent tallies. But the survey is sponsored, its key results are opinions, and the four-month patch total doesn't match other published counts. The question Manfra ends on is fair: is your technology partner helping you build, or keeping you on a treadmill? Ask it of every vendor you use, including the one writing the op-ed.

 

References

  1. When the legacy software meant to deliver AI becomes a trap - Federal News Network Federal News Network 2026-11-19T21:24:49+00:00
  2. New survey finds enterprises trapped between legacy software and AI modernization cloud.google.com
  3. Microsoft's Patch Tuesday updates: Keeping up with the latest fixes – Computerworld computerworld.com