Copilot promises trust failures and questions about hiring and layoffs
Microsoft has built its artificial intelligence narrative around a major promise: a digital assistant that increases productivity, understands the context of work and connects language models to business data within a secure environment. Satya Nadella, its chairman and chief executive, has been the leading voice presenting this transformation as a new phase in computing.
Testing that narrative requires looking beyond promotional demonstrations and adoption figures. It requires examining output accuracy, data protection, transparency about failures and the management of the workforce building these products.
The sources establish specific security failures, government criticism of delayed corrections to inaccurate public information, and substantial layoffs alongside continued use of H-1B petitions. They do not establish that all Nadella’s statements are false, that every Copilot product has failed, or that Microsoft operates a general policy favoring employees of a particular nationality. The strongest accountability case starts with what can be demonstrated and identifies what remains undisclosed.
1 The promise of a trusted assistant
When Microsoft announced Microsoft 365 Copilot on 16 March 2023, Nadella said it would mark a major step in the evolution of human interaction with computing, fundamentally change work and unlock a new wave of productivity growth. The announcement described Copilot as connected to organizational data and context, delivering accurate and relevant responses while inheriting existing security, compliance and privacy policies. The proposition combined capability with trust: access to business data would make the assistant useful, and enterprise controls would make it suitable for professional use. [1]
These propositions require separate tests. Producing a draft quickly does not establish that its contents are correct. Accessing a document does not guarantee sound interpretation. Inheriting a user’s permissions does not, by itself, prevent an attacker from manipulating an assistant into mishandling information within those permissions.
The question for leadership is how much trust the actual testing supports, and whether customers understand the limits before relying on the product.
2 What the productivity studies actually measured
In November 2023, Microsoft reported that 70 percent of surveyed Copilot users felt more productive. The survey covered 297 users in an early access program. It measured the experiences and perceptions of a particular group; it did not establish that every user’s productivity increased by 70 percent. [2]
In a separate experiment involving 147 participants, Microsoft reported that Copilot users completed a series of searching, summarizing and writing tasks 29 percent faster, with no statistically significant difference in accuracy between groups. [2]
These are findings worth reporting within their study conditions. They do not, alone, demonstrate an equivalent improvement in an entire organization’s profits, decision quality or productivity after training, verification and correction are included.
The issue requiring scrutiny is the distance between a bounded research result and a broad marketing message. Establishing that a specific statement is misleading requires identifying its wording, what the study measured and whether omitted conditions materially changed the audience’s understanding.
3 An inaccurate explanation remained public for months
The US Cyber Safety Review Board’s review of the summer 2023 Microsoft Exchange Online intrusion concluded that Microsoft’s security culture was inadequate and identified avoidable errors contributing to the intrusion.
The report recorded a particularly significant disclosure issue. Microsoft acknowledged to the Board in November 2023 that its 6 September blog post about the intrusion’s root cause was inaccurate. It did not update the post until 12 March 2024, after repeated Board questions about its plans to issue a correction. [3]
This was more than a technical problem. It raises a direct question about disclosure: why did an inaccurate explanation remain public after the company acknowledged the problem to investigators?
The episode justifies questioning Nadella as the company’s leader. It does not, by itself, establish that he wrote the explanation or ordered the delay. Personal responsibility for that decision would require evidence showing when he was informed, what he was told and what instructions he issued.
4 Security above all else as a measurable commitment
On 3 May 2024, Nadella declared that security should take priority over competing objectives, including the release of new features. He announced principles covering secure design, secure defaults and secure operations, and said part of senior leadership compensation would depend on progress against security plans and milestones. [4]
That commitment can be assessed through actual decisions. Which features were delayed because of security risks? Which failures affected leadership assessments and compensation? Did corrections and disclosures become faster? Can customers verify the results of remediation?
A later vulnerability does not automatically prove that the commitment was dishonest. Repeated failures nevertheless require evidence that the stated priority has changed decision making.
5 EchoLeak and the boundary between external content and internal data
A technical case study documented EchoLeak, CVE-2025-32711, in Microsoft 365 Copilot. It demonstrated how a specially crafted email could exploit prompt injection to exfiltrate information from the assistant’s context without the user interacting with the malicious message. The significance was the failure of a trust boundary between untrusted external material and internal information accessible to the assistant. [5]
Reporting that cited Microsoft’s disclosure said the vulnerability was fixed on the server side in May 2025 and that there was no evidence of exploitation in the wild. This supports describing a serious demonstrated vulnerability that was remediated. It does not support claiming that all customers’ data was stolen. [6]
The accountability question is how this path across trust boundaries escaped pre-release testing, and what evidence shows that related failure modes have since been addressed.
6 Confidential email and behavior that breached the intended design
In February 2026, Microsoft acknowledged an issue in which Copilot Chat could return content from emails labeled confidential that had been authored by the user and stored in Drafts and Sent Items.
Microsoft said the problem did not give anyone access to information they were not already authorized to see. It nevertheless acknowledged that the behavior did not match the intended experience, which excluded protected content from Copilot access, and said a configuration update had been deployed. [7]
The reported incident did not establish disclosure to an unauthorized person. It established a failure in the expected handling of protected content. That distinction matters: whether a user may read a document and whether an assistant should process it are separate questions.
7 The cost of checking the assistant
Microsoft’s guidance warns that Copilot output can look polished while being wrong or incomplete, and that users remain responsible for validating it before acting. It also says Copilot can help review its output but cannot certify its own correctness. [8, 9]
Those warnings are necessary. They also raise an economic question the evidence presented here does not settle: how much productivity remains after source checking, correction and rework are included?
A useful tool may still require review. The business benefit should be measured after that review cost, particularly when the output affects consequential decisions.
8 Why the evidence does not establish universal Copilot failure
These incidents establish specific failures. They do not establish that every Copilot product or use case has failed.
In its 2025 annual report, Microsoft announced more than 100 million monthly active users across its commercial and consumer Copilot family. This is a company-reported adoption metric. It does not mean 100 million paying Microsoft 365 Copilot subscribers, and it is not an independent measure of accuracy or safety. [10]
Reach and usage require different evidence from reliability. Each incident should also be attributed to the affected product rather than treating the shared Copilot brand as a single system.
9 Layoffs and Nadella’s acknowledgement of the apparent contradiction
In a message dated 24 July 2025, Nadella acknowledged that Microsoft was thriving and increasing investment while also conducting layoffs. He said overall headcount was relatively unchanged and described the situation as an “enigma of success.” [11]
Associated Press reported approximately 9,000 worldwide job cuts in July 2025, following around 6,000 in May and additional reductions in June. These were global figures. They cannot all be described as American workers without a geographic and citizenship breakdown. [12]
In July 2026, Microsoft announced approximately 4,800 further role eliminations. It said it had redeployed more than 4,000 employees into new roles over the preceding year. That is relevant evidence of internal movement, but it does not disclose what share of employees facing elimination were redeployed or how their skills compared with roles for which recruitment continued. [13]
Stable total headcount can coexist with major changes in who is employed, which skills are retained, where work is performed and how it is compensated. Those changes require a more detailed account than the overall headcount.
10 Amazon leads the comparison for new H-1B employment
Microsoft was not the largest recipient of H-1B approvals for initial employment in the National Foundation for American Policy’s analysis of fiscal year 2025 USCIS data. Amazon ranked first, followed by Meta, Microsoft and Google. [14]
| Rank | Company | Initial employment approvals FY2025 |
| 1 | Amazon | 4,644 |
| 2 | Meta Platforms | 1,555 |
| 3 | Microsoft | 1,394 |
| 4 | 1,050 |
These are initial-employment petition approvals, not all renewals, the total active H-1B workforce or a count of people arriving from abroad. Beneficiaries may already be in the United States. The ranking corrects the earlier suggestion that Microsoft led this measure. [14]
11 Whether laid-off workers were replaced remains a job-level question
In September 2025, Senators Chuck Grassley and Dick Durbin questioned Microsoft as part of scrutiny of major employers conducting layoffs while filing H-1B petitions. Their questions covered displacement of Americans, comparable compensation, domestic recruitment and the use of contractors. These were oversight questions, not judicial findings establishing replacement. [15, 16]
In an 8 October 2026 statement, Microsoft said 80 percent of approximately 6,000 H-1B filings in the preceding fiscal year concerned extensions or status changes for existing employees. It said filings for new employees involved people already legally in the United States and equaled about one percent of its US workforce. [17]
That defense undermines treating the total filing count as the number of newly hired replacements. It does not settle whether individual eliminated roles were followed by substantially similar roles filled by new workers, or whether affected employees received a fair opportunity to transfer.
Establishing replacement requires comparing duties, skills, location, level and timing. Placing a layoff total beside a petition total does not establish the causal connection. Nor should H-1B and permanent residence filings simply be added together as separate workers: the same person can appear in both processes.
12 The announced October 2026 action was not a final finding
On 8 October 2026, the US administration announced a suspension of the acceptance and processing of Permanent Labor Certification Program applications involving companies including Microsoft, accompanied by allegations of immigration-system abuse. Reuters also reported Microsoft’s defense concerning existing employees and comparable pay. The announcement and allegations do not, by themselves, establish the alleged violations. [18]
The announced action concerned PERM. It did not mean cancellation of existing H-1B visas or suspension of the entire H-1B program for Microsoft. This account describes the announced action; it does not establish the outcome of any subsequent proceedings. [19]
13 Official employment discrimination settlements require precise attribution
In September 2020, the US Department of Labor announced a settlement under which Microsoft would pay three million dollars in back pay and interest to 1,229 applicants. The Department identified alleged hiring disparities affecting Asian applicants in specified roles and Black and Hispanic applicants in software engineering roles. Microsoft did not admit liability. [20]
In December 2021, the Justice Department announced a separate settlement concerning unnecessary requests for specific immigration documents from noncitizens, including lawful permanent residents, and procedures involving people who did not require employer visa sponsorship to work. [21]
These precedents justify examining hiring controls. They do not establish favoritism toward Indian nationals or H-1B workers. Some affected applicants were Asian or were noncitizens authorized to work without sponsorship. The periods examined also varied; every underlying event cannot automatically be attributed to a decision by Nadella.
14 Nationality is not evidence of control or favoritism
The reviewed sources do not provide a workforce census establishing that most Microsoft employees are Indian nationals or hold H-1B status. An employee’s ancestry does not determine current citizenship or immigration status. Microsoft’s corporate filing identifies it as incorporated in Washington State and headquartered in Redmond. [22]
The testable issue is the fairness of decisions about hiring, promotion and dismissal. The Equal Employment Opportunity Commission explains that preferring foreign workers, including workers with a particular visa status, over Americans can constitute national-origin discrimination depending on the facts. That is a general standard, not a Microsoft-specific finding in the material reviewed here. [23]
Evidence of favoritism would require decisions, instructions, comparative records or corroborated testimony. The names or ancestry of managers do not supply that proof.
15 Questions proposed for a right of reply
The following questions are proposed. They were not sent as part of preparing this report.
1. When was Nadella informed that the published explanation of the 2023 intrusion was inaccurate, and why was correction delayed?
2. Which features were postponed to implement the commitment to put security first?
3. How did security failures affect leadership evaluations and compensation?
4. Will Microsoft publish product-specific and language-specific Copilot error rates using an independently reviewable methodology?
5. What productivity gains remain after verification, correction and rework?
6. How did EchoLeak and the confidential-email issue escape pre-release testing?
7. How many employees actually work in H-1B status after duplicate filings are excluded?
8. How many US roles were eliminated and followed by substantially similar filled positions within six months?
9. How many employees facing layoffs applied for internal positions, how many were accepted, and why were others rejected?
10. Will Microsoft publish an independent assessment of compensation and promotion decisions controlling for role, experience, location and immigration status?
11. How much work moved to contractors inside or outside the United States after layoffs?
12. Will Microsoft publish its full responses to oversight questions and the findings of related reviews?
The accountability case supported by the evidence
The responsibility this report places on Nadella is specific: provide verifiable evidence that promises about trust, security and fair employment are reflected in decisions and outcomes. The documents establish failures warranting scrutiny. Allegations of personal intent, systematic replacement or nationality-based favoritism require further evidence.
Scope note: This report consolidates the publicly sourced investigation in the preceding discussion. It does not independently authenticate private incident records or correspondence, and it does not present those materials as verified evidence.
Source register
[1] Microsoft. Introducing Microsoft 365 Copilot. 16 March 2023. https://blogs.microsoft.com/blog/2023/03/16/introducing-microsoft-365-copilot-your-copilot-for-work/
[2] Microsoft WorkLab. What Can Copilot’s Earliest Users Teach Us About Generative AI at Work. November 2023. https://www.microsoft.com/en-us/worklab/work-trend-index/copilots-earliest-users-teach-us-about-generative-ai-at-work/
[3] US Cyber Safety Review Board. Review of the Summer 2023 Microsoft Exchange Online Intrusion. 2024. https://www.cisa.gov/sites/default/files/2024-03/CSRB%20Review%20of%20the%20Summer%202023%20MEO%20Intrusion%20Final_508c.pdf
[4] Satya Nadella Microsoft. Prioritizing security above all else. 3 May 2024. https://blogs.microsoft.com/blog/2024/05/03/prioritizing-security-above-all-else/
[5] AAAI Symposium Series. EchoLeak technical case study. 23 November 2025. https://ojs.aaai.org/index.php/AAAI-SS/article/view/36899
[6] BleepingComputer. Zero-click AI data leak flaw uncovered in Microsoft 365 Copilot. 2025. https://www.bleepingcomputer.com/news/security/zero-click-ai-data-leak-flaw-uncovered-in-microsoft-365-copilot/
[7] BleepingComputer. Microsoft says bug causes Copilot to summarize confidential emails. February 2026. https://www.bleepingcomputer.com/news/microsoft/microsoft-says-bug-causes-copilot-to-summarize-confidential-emails/
[8] Microsoft Support. Validate Copilot output before you act on it. Undated guidance. https://support.microsoft.com/en-gb/microsoft-365-copilot/validate-copilot-output
[9] Microsoft Support. Validate Copilot output before you act on it US edition. Undated guidance. https://support.microsoft.com/en-us/microsoft-365-copilot/validate-copilot-output
[10] Microsoft. Annual Report 2025. 2025. https://www.microsoft.com/investor/reports/ar25/
[11] Satya Nadella Microsoft. Recommitting to our why what and how. 24 July 2025. https://blogs.microsoft.com/blog/2025/07/24/recommitting-to-our-why-what-and-how/
[12] Associated Press. Microsoft’s largest layoff in years hits Xbox sales and other divisions. July 2025. https://apnews.com/article/f44079957b12370f72e24edebe9fcc6b
[13] Microsoft. The latest in our company transformation. 6 July 2026. https://blogs.microsoft.com/blog/2026/07/06/the-latest-in-our-company-transformation/
[14] National Foundation for American Policy. H-1B Petitions and Denial Rates FY 2025 release. November 2025. https://nfap.com/wp-content/uploads/2025/11/H-1B-Petitions-and-Denial-Rates-FY-2025.DAY-OF-RELEASE.2025.pdf
[15] US Senate Judiciary Committee. Grassley Durbin scrutiny of employers using H-1B. 25 September 2025. https://www.judiciary.senate.gov/press/rep/releases/grassley-durbin-take-aim-at-tech-finance-and-retail-sectors-for-favoring-h-1b-visa-holders-over-american-workers
[16] Senators Grassley and Durbin. Letter to Microsoft on H-1B. 24 September 2025. https://www.grassley.senate.gov/download/grassley-durbin-to-microsoft_-h-1b?download=1
[17] Microsoft. Comment on Recent Immigration Questions. 8 October 2026. https://blogs.microsoft.com/on-the-issues/2026/10/08/microsoft-comment-on-recent-immigration-questions/
[18] Reuters. Trump administration freezes green cards for Microsoft IT firms probes universities. 8 October 2026. https://www.reuters.com/business/us-suspending-permanent-residency-program-for-microsoft-vance-says-2026-10-08/
[19] Axios. H-1B visas are not suspended What Vance is targeting. October 2026. https://www.axios.com/2026/10/08/h1b-vs-perm-vance-tech-worker-ban
[20] US Department of Labor. Microsoft agreement resolving alleged hiring discrimination affecting 1229 applicants. 18 September 2020. https://www.dol.gov/newsroom/releases/ofccp/ofccp20200918
[21] US Department of Justice. Settlement with Microsoft resolving immigration-related discrimination claims. 7 December 2021. https://www.justice.gov/archives/opa/pr/justice-department-settles-microsoft-resolve-immigration-related-discrimination-claims
[22] Microsoft SEC filing. Form 10-K for year ended 30 June 2025. 2025. https://www.sec.gov/Archives/edgar/data/789019/000095017025100235/msft-20250630.htm
[23] US Equal Employment Opportunity Commission. Discrimination Against American Workers Is Against The Law. Undated guidance. https://www.eeoc.gov/discrimination-against-american-workers-against-law
