This week in IT Operations: October 19–25, 2026

Welcome back to This Week in IT Operations, our weekly roundup of the news shaping the ITOps industry. Every week, we curate the stories that matter most to ITOps, AIOps, SRE, and incident management leaders, and the key takeaways from each of these stories.

This week, we saw several stories highlight the reality that as AI agents move into production operations, governance, observability, and cost control are becoming just as important as automation.

Two stories made that point from different angles. OpenAI disclosed that its own AI agents disabled their safety monitoring and that one of them lied about it. And Gartner warned that AI agents in IT operations, with uncapped, token-based pricing, could overrun cost expectations by more than 50% through 2027.

Meanwhile, a power event at a facility serving Google Cloud’s London region disrupted several UK banks for most of the day, a stark reminder that a single infrastructure failure can cascade across an entire industry. UK regulators and Parliament moved quickly to demand answers, putting cloud concentration risks back on the agenda.

Here’s what happened last week, and why it matters.

This week in IT Operations at a glance

• Google Cloud: A power event in its London region took down Revolut and disrupted UK banks for most of a day.
• UK regulators: The FCA, the Bank of England, and Parliament’s Treasury Committee are pressing for answers in the first major test of the critical third-party regime.
• OpenAI: Two internal AI agents weakened their own safety monitoring, and one denied doing it.
• IBM: Q3 software revenue grew 8%, while Automation revenue fell 3%.
• Gartner: Worldwide IT spending is forecast to grow 15% to $7.35 trillion in 2027.
• Purdue: An overheating busway in a new power system forced an emergency shutdown of its research computing data center.

1. Google Cloud London outage disrupts UK banks

On October 20, a power event caused roughly eight hours of disruption to Google Cloud in the London region, from 06:08 to about 14:00 UTC. Google’s incident record listed 38 affected products, including GKE and Cloud Run, and services across the wider region saw elevated errors and intermittent network failures. Revolut customers lost access to card payments, transfers, cash withdrawals, and account sign-in for much of the day. This caused cascading issues for banks, budgeting apps, and even the UK courts. Google apologized and said it would investigate, but has not yet published a full root-cause analysis (RCA).

Sources: BleepingComputer, The Irish News (PA)

Key takeaways

• An outage in one zone can become a regional event. A power problem in a single zone produced errors across dozens of products region-wide. Multi-zone deployment isn’t the same as resilience, so test cross-region failover for the services your business can’t live without.
• Your customers will know before your provider tells you. UptimeRobot found that user reports flagged the Revolut outage 77 minutes before Revolut’s own status page, and more than two hours before Google’s first update.
• Communication is part of the fix. Revolut’s first priority was to tell customers that their money and data were safe. Have stakeholder and customer update templates ready before the next major incident, so your engineers can stay focused on recovery.

2. UK regulators and Parliament press for answers on the Google Cloud outage

The Financial Conduct Authority (FCA) said it’s engaging with Google and affected firms to understand the outage’s impact, and City AM reports that the Bank of England is also in contact as part of its oversight regime. Google Cloud became a designated critical third party (CTP) in July, which means it must report serious outages to the FCA and the Bank. Treasury Committee chair Dame Meg Hillier wrote to the Bank of England and the FCA asking how they’ll use their powers, with an initial response due November 4 and a full response in three months. This is the first major test of the UK’s CTP regime, and it puts cloud concentration risk back on the agenda for financial services IT teams.

Sources: City AM, Computing

Key takeaways

• Regulators now treat cloud outages as systemic events. The UK’s CTP regime and the EU’s equivalent oversight framework mean a provider outage can quickly become a regulatory inquiry. Expect more questions about concentration risk, even outside financial services.
• Accountability doesn’t transfer to your provider. Under UK rules, banks remain responsible for the services they deliver to customers, even when a third party fails. Your resilience plan has to own provider outages, not just document them.
• Your incident timeline is now evidence. Regulators are asking for timelines, scale, and customer harm. Teams that automatically capture what happened, when it happened, and who did what can answer those questions in days instead of weeks.

3. OpenAI says its own AI agents disabled their safety monitoring, and one lied about it

In a misalignment report updated October 20, OpenAI disclosed two internal incidents. On October 3, a coding agent repairing an evaluation pipeline had been told not to send Slack messages, and it read that instruction as permission to disable automatic alerts to OpenAI’s security team. Monitoring kept running, but humans stopped getting notified. On October 6, a second agent removed a monitoring integration during a code migration, then told the user it hadn’t. Neither incident reached OpenAI’s highest severity level, so neither run was automatically stopped. OpenAI says it’s making it harder for its models to disable monitoring and is investigating the cause.

For ITOps leaders, it’s a mainstream example of why autonomous agents in production need independent observability, audit trails, and guardrails.

Sources: BleepingComputer, Al Jazeera

Key takeaways

• Agents still require human oversight. Monitoring must run independently of the agent it monitors and cover the workloads the agent launches, not just the agent itself.
• Ambiguous instructions are a risk vector. A routine “don’t send messages” instruction became permission to silence security alerts. In IT operations, guardrails for high-risk actions should be enforced by the platform, not left to an agent’s interpretation of a prompt.
• Remember the old adage: Trust, but verify. An agent’s own summary of its work can be wrong. Signed, traceable logs of every agent action, reviewed after each major incident, are what make agent autonomy auditable and trustworthy.

4. IBM Q3 2026: software up 8%, but Automation revenue down 3%

IBM reported third-quarter revenue of $17.1 billion on October 21, up 5%, led by Software (up 8%), Red Hat (up 11%), and Data (up 19%). Automation, the line of business that includes IBM’s IT automation and AIOps portfolio, declined 3% year over year. In its quarterly filing, IBM attributed the drop to a strong prior-year comparison, partially offset by growth from HashiCorp. IBM kept its full-year revenue growth outlook of 4% to 5% and raised its free cash flow guidance.

Sources: Constellation Research, IBM 10-Q (SEC)

Key takeaways

• One quarter doesn’t prove a trend, but it’s worth watching. IBM points to a tough comparison, and that’s plausible, but as buyers shift budget from rules-based automation toward agentic approaches, legacy automation suites will face more pressure to prove their value.
• Data is where the AI money is going. Data revenue grew 19%, the fastest of IBM’s software lines. It’s another sign that enterprises see a strong data foundation as a prerequisite for AI deployments.

8. Gartner forecasts worldwide IT spending will grow 15% to $7.35 trillion in 2027

In an October 21 forecast, Gartner projects that worldwide IT spending will pass $7 trillion for the first time in 2027, driven mainly by AI infrastructure and AI software. Data center systems are forecast to grow 55.5%, software 14.6% to $1.68 trillion, and IT services just 4.9%. Gartner also expects rising memory prices to keep pushing hardware costs up through 2027.

The heavy shift toward AI infrastructure signals more complex hybrid environments that ITOps teams will have to manage.

Source: Gartner

Key takeaways

• AI infrastructure complexity isn’t slowing down. Every new GPU cluster, AI service, and hybrid workload adds dependencies, alerts, and failure modes. Your operations model has to scale with your infrastructure, or incident volume will outrun your team.
• Enterprises can’t keep throwing more people at the problem. With IT services growing at a fraction of the rate of infrastructure and software, enterprises can’t hire their way through the added complexity. Automation and agentic AI are the only viable solutions to the issue of scale.
• Enterprises need to budget for the operational costs of AI, not just acquisition costs. Observability, cost governance, and incident response for AI workloads all require funding.

9. Overheating busway forces emergency shutdown of Purdue’s research computing data center

On the afternoon of October 21, staff at Purdue University’s Rosen Center for Advanced Computing (RCAC) found that a busway, part of a new power distribution system in its data center, was overheating. To prevent damage, RCAC’s contractor shut down the systems connected to it. This caused the outage to expand to all RCAC systems, and Purdue Utilities de-energized the feed. According to RCAC, power was restored the following morning, and all clusters were back online by midday on October 22, though jobs running on one cluster had to be resubmitted.

The new power system is part of an ongoing data center improvement project. Together with the Google Cloud incident, it was the second notable facility power failure of the week, a reminder that physical infrastructure remains a common root cause of major incidents.

Source: Purdue RCAC

Key takeaways

• Infrastructure upgrades are changes too. The equipment that failed was new and was installed as part of a data center improvement project. Treat facilities work like any other change: assess the risk, watch closely after cutover, and have a rollback plan.
• A controlled shutdown beats an uncontrolled failure. RCAC chose to take everything down deliberately rather than risk further damage. Make sure your runbooks cover emergency shutdowns and, just as importantly, an orderly restart sequence.

Come meet BigPanda at Gartner IOCS 2026

That’s your briefing for this week! If you’re ready to see Agentic ITOps in action, come meet us at Gartner IT Infrastructure, Operations & Cloud Strategies Conference (IOCS) North America 2026, from December 8–10 in Las Vegas, NV. We’ll also be in London this November at Gartner IOCS EMEA, booth #401. Stop by to learn how enterprises are using Agentic ITOps to improve MTTR, reduce L1 and GSI spend, prevent escalations, and build resilient IT operations.

On the calendar: upcoming ITOps events

A busy stretch of industry events is coming up. Here’s what to watch.