The week of Sept. 14-18: What happened, what matters, what's next

By Madeleine Streets , senior editor at InformationWeek

As enterprises expand their use of increasingly powerful technology, IT leaders must navigate a series of constraints that extend beyond the systems themselves. This week's news highlighted several of those challenges, from the power needed to support growing computing demand to questions about who controls AI systems and how much organizations can rely on complex cloud infrastructure.

Those issues surfaced in different ways this week. Microsoft proposed new rules to keep AI under human control, while a brief use of a Chinese AI model on a U.S. government website highlighted the geopolitical considerations that can accompany model choices. Elsewhere, major technology companies turned to the power grid to support additional data center capacity, and a Salesforce outage served as a reminder of the dependencies built into cloud services.

Here are four key developments worth watching this week:

News Highlights Sept. 14-18

Microsoft proposes code of conduct for keeping AI under human control

What happened: On Sept. 14, Microsoft unveiled a draft code of conduct for its AI systems that would require them to accept correction, comply with shutdown requests and communicate their behavior in understandable ways. The company is seeking public feedback on the framework for six weeks. Microsoft AI CEO Mustafa Suleyman described it as a kind of "constitution" for future AI models, that follows a set of design principles that the company has termed "Humanist AI."

Why it matters: The proposal, published amidst high-profile calls from tech CEOs to slow down AI development, puts some concrete boundaries around what responsible behavior could look like as AI systems become more capable and autonomous. For enterprises, those expectations could increasingly shape how AI systems are evaluated and deployed, particularly as they are given greater control over business processes. It also provides a more tangible way for IT leaders to consider whether an AI system can be trusted to remain within defined boundaries.

Google, Nvidia and Anthropic turn to the grid to unlock more data center capacity

What happened: Google, Nvidia and grid software company Emerald AI launched the AI Energy Management Alliance onSept. 16, with Anthropic and several utilities also joining. The group is promoting demand-response technology that can temporarily reduce or shift data center workloads when electricity demand is high. The companies said the approach could eventually allow up to 100 gigawatts of additional data center capacity to connect to the grid.

Why it matters: The availability of electricity is becoming a defining factor in how quickly new computing capacity can come online. The alliance points to a different way of addressing that constraint, by making workloads themselves more flexible. That could give enterprises another consideration when designing infrastructure for compute-intensive workloads: whether some workloads can be paused, shifted or scheduled around the availability and cost of power.

U.S. government briefly uses Alibaba's Qwen model on Federal Register site

What happened: The National Archives briefly used Alibaba's Qwen AI model to help users search proposed federal regulations. The feature was taken down after social media users flagged its use, amid heightened U.S.-China tensions over AI technology and recent FBI allegations that Alibaba copied some of its technology from Anthropic. Alibaba and Chinese officials have rejected those allegations.

Why it matters: The reaction reflects the sensitivity around Chinese AI technology in the U.S., particularly when government systems are involved. Qwen is an open-weight model that can be run on an organization's own infrastructure, enabling it to keep data under its control. In this case, the Federal Register contains public information, but the incident highlights the competing considerations around using Chinese-developed models even when their technical architecture offers data-control advantages.

Salesforce outage exposes risks in cloud dependencies

What happened: Salesforce experienced a roughly seven-and-a-half-hour outage on Sept. 16, which was also Day 2 of its annual Dreamforce user conference. The outage affected multiple instances across regions, causing access problems, severe delays and intermittent errors. Salesforce initially attributed the incident to an external dependency failure involving a legacy login server; the company subsequently applied fixes and restored service around 3 p.m. EDT.

Why it matters: The outage showed how a failure in a relatively narrow component can affect access to a much broader cloud platform. As enterprises rely on cloud services for increasingly critical business functions, understanding those dependencies becomes an important part of resilience planning. It also underscores why application availability depends on more than the status of the application itself.

From InformationWeek: Worth the read

You'll find a wealth of insight and analysis on the InformationWeek site, but here are a few curated picks from the week that should be at the top of your reading list.

AI panic is giving CIOs a new trust problem This week's Big Tech Story explores the current panic over frontier AI safety and how CIOS can instill organizational confidence in enterprise AI roadmaps amid these fears.

Altman, Amodei talk AI pacing at Dreamforce Contributing writer Myles Suer reports from San Francisco on how the CEOs of OpenAI and Anthropic both addressed the push to slow down AI in two separate sessions.

InformationWeek Podcast: Does AI help or trigger alert fatigue? In his latest podcast episode, senior editor Joao-Pierre Ruth is joined by CISOs at Tricentis and ABBYY to discuss how AI can be part of a layered security strategy.

The multi-AI model stack is here. Now someone has to manage it Contributing writer Stephanie Overby explores how CIOs can manage and govern multi-model stacks, especially as those models and costs shift over time.

Coming up: Dates to watch

Sept. 22–24: Okta's annual Oktane event gets underway in Las Vegas, with the opening keynote expected to highlight developments in identity and AI security. Identity management for AI agents and other non-human identities will be worth watching.

Sept. 23–24: Meta's developer event Meta Connect 2026 will showcase developments in AI, AI glasses and virtual reality. With wearable AI increasingly moving toward workplace applications, there could be announcements on how these devices connect to enterprise systems.

Sept. 24: President Donald Trump and Chinese President Xi Jinping are scheduled to meet in Washington, with technology restrictions and AI among the issues expected to feature in broader discussions on trade and the U.S.-China relationship.

From AI systems and data centers to cloud dependencies and geopolitical tensions, this week's news showed how many factors can shape enterprise technology decisions beyond the technology itself. What external constraints are having the biggest impact on your IT strategy right now? Let us know at editors@informationweek.com.