Microsoft Workers’ AI Open Letter: Mandatory Use, Job Cuts, Contracts and Environmental Safeguards

The most striking allegation concerns internal planning language about making customers dependent on AI. But the broader dispute is about accountability: who decides where AI belongs, who measures whether it helps, and who bears the consequences when it does not?

There are important limits to the reporting. Rock Paper Shotgun said the organizers’ identities and the number of signatories were unclear when its article appeared. It had requested Microsoft’s comment. The letter should therefore be understood as a group’s stated position—not a workforce-wide verdict, and not independent proof of every allegation it contains.

Mandatory AI use meets a product-quality complaint​

According to Rock Paper Shotgun’s account, the workers describe deteriorating morale alongside heavy workloads, reorganizations, shifting priorities and fears of replacement or layoffs. They allege that employees are expected to use AI across much of their work without adequate training or a convincing explanation beyond faster delivery.

Their complaint is not simply that learning new software takes effort. It is that the claimed efficiency gains can disappear into correcting generated material, adapting to changing tools and repairing or reversing poor-quality features.

The workers’ requested changes are consequently more specific than a blanket rejection of generative AI:

• Give employees a meaningful role in AI product-strategy decisions.
• Incorporate customer feedback and expertise from different development disciplines.
• Provide proper training rather than treating adoption as self-explanatory.
• Stop linking AI usage to rewards, punishment or workforce reductions.
• Establish protections against what the writers characterize as unfair AI-driven job cuts.

Rock Paper Shotgun places the letter against a backdrop of voluntary buyouts and earlier layoffs. That context helps explain the anxiety, but it does not establish that AI caused each workforce decision or that the letter represents everyone affected.

For enterprise IT leaders, the practical lesson is a measurement problem: tool usage and useful work are not the same thing. As an analytical recommendation—not a finding about Microsoft’s internal results—an AI rollout should measure completed work, correction effort, quality and employee feedback. Counting prompts alone risks turning an adoption dashboard into a very expensive applause meter.

What the addiction allegation does—and does not—establish​

The workers reportedly cite internal planning documents containing language about making people dependent on AI, alongside unnamed research that they say raises concerns about agent safety.

There is relevant earlier reporting. On June 2, 2026, 404 Media reported that planning documents for Microsoft’s Scout assistant used the phrase “make people addicted” and described adding functionality afterward. That provides a concrete antecedent for the controversy, although it does not establish that the workers’ letter refers to precisely the same document.

There is also a material counterpoint. Windows Central reported that Satya Nadella rejected addiction as an objective, while a Microsoft spokesperson said Scout was intended to help people complete tasks rather than encourage dependency or increase screen time. That earlier response is relevant context, not a response to this new letter.

The distinction matters. Evidence that troubling language appeared in a planning document is not equivalent to proof of an approved, company-wide policy. Equally, management’s rejection of that language does not answer every question about how individual teams set engagement targets.

The workers’ broader safety claim remains harder to assess. Rock Paper Shotgun does not identify the research, explain its methods or provide enough detail to evaluate its conclusions. Their demands include expanding Responsible AI and Sensitive Use teams and improving communication about potential harms to customers.

Government contracts are central, not incidental​

The letter also challenges Microsoft’s relationships with Israel’s military and U.S. Immigration and Customs Enforcement. As reported by Rock Paper Shotgun, the writers argue that supplying cloud and AI capabilities can expose Microsoft to involvement in surveillance, detention, deportation and other human-rights abuses.

They seek disclosure of government contracts, including relationships involving contractors, police and armed forces. They also want contracts terminated where organizations are directly or indirectly involved in human-rights violations.

These demands sit alongside a significant documented Microsoft decision. On September 25, 2025, Microsoft president Brad Smith announced that the company had disabled specified subscriptions and services for a unit within Israel’s Ministry of Defense. Microsoft said its review had found evidence supporting elements of reporting about Azure storage consumption and AI-service use.

That was a limited action, not an announcement ending all Israeli defense relationships. Smith explicitly said other cybersecurity work for Israel and countries in the region would continue. Microsoft also stated that its terms prohibit using its technology for mass surveillance of civilians. The distinction leaves a substantive question behind the workers’ demands: which services remain available, under what restrictions, and with what oversight?

The Gaza context also requires precise attribution. On September 16, 2025, the UN Independent International Commission of Inquiry concluded that Israel had committed genocide against Palestinians in Gaza. That is the commission’s finding; it does not itself establish Microsoft’s legal responsibility for the conduct alleged in the workers’ letter.

The reporting on the letter does not identify specific ICE workloads or establish how particular Microsoft AI services were used in particular enforcement actions. Those gaps should not be filled by assumption.

Data-center promises face a different test​

Environmental concerns form another substantial part of the letter. The workers criticize water consumption, emissions and community impacts associated with AI infrastructure, and ask Microsoft to reaffirm its 2030 carbon-negative, water-positive and zero-waste goals.

They also seek stronger engagement with regulators, avoidance of resource-scarce locations, protection for local workers and respect for unionized construction trades. Their argument is that construction activity should not be confused with a permanent employment settlement for a community.

Microsoft’s published Community-First AI Infrastructure plan provides a useful comparison. Announced on January 13, 2026, it includes commitments covering electricity costs, water replenishment, local employment, taxes and community training. It also describes a partnership with North America’s Building Trades Unions and acknowledges that construction typically employs thousands while operations employ hundreds. These are stated commitments and expectations, not proof that every project has delivered them.

The disagreement is therefore not whether Microsoft has published community commitments—it has—but whether those commitments adequately address the harms and trade-offs the workers identify.

Financial exposure completes the argument​

Finally, the workers criticize interconnected investments among AI companies and seek reduced dependence on OpenAI and other partners. They want more transparency about investments and argue that concentrated relationships can distribute risk to employees, suppliers, customers and shareholders.

Rock Paper Shotgun also reports their concern about AI-linked companies’ prominence in stock-market indexes, but provides no underlying calculation. That should remain an attributed risk argument, not a numerical finding or a prediction of impending collapse.

Taken together, the letter is less a demand to abandon AI than a demand to change who governs its deployment. Its authors want worker participation, customer benefit, enforceable safeguards and community protections to carry more weight than deployment speed.

For Windows users, developers and enterprise administrators, that is the substantive issue worth watching. The useful question is not whether an organization can put AI into another workflow. It is whether it can demonstrate that doing so improves the work—and accept responsibility when the answer is no.