Meta and Microsoft are reportedly reducing employees’ use of Anthropic’s Claude while encouraging company-backed alternatives. Microsoft has lowered its projected internal Anthropic spending by roughly one-third, while Meta’s Claude Code user count has fallen from about 60,000 to around 30,000.
The figures originate in The Information’s October 5, 2026 report, which relies on unnamed sources. Accessible accounts from PYMNTS and The Decoder describe a push toward GitHub Copilot and OpenAI models at Microsoft, and Muse Code and MetaCode at Meta.
Neither change establishes a confirmed split with Anthropic. Microsoft’s reported response leaves engineers free to choose other models, and moving work into Copilot does not necessarily eliminate Claude from the underlying workflow.
The more consequential question is whether Anthropic can keep expanding inside customers that also have strategic reasons to promote competing products. Model quality may win employee enthusiasm without determining where corporate budgets go.
Microsoft Reportedly Cuts Its Anthropic Spending Forecast
According to PYMNTS’ account, Microsoft had been on track to spend approximately $1 billion during 2026 on internal use of Anthropic’s technology. That projection has since fallen by roughly one-third.
This is a reported revision to an expected annual bill, not an audited disclosure that Microsoft canceled a particular contract or recovered money already spent. The accounts do not provide enough detail to establish how much of the change comes from fewer users, lower consumption, different model choices, or tighter budgets.
The spending covers several uses of Anthropic technology, according to PYMNTS, including Claude Code and Claude models within Microsoft’s Copilot. It should not be treated as a Claude Code-only budget.
The Decoder’s coverage says Microsoft executives directed employees toward GitHub Copilot and OpenAI models. Those alternatives belong to different categories: Copilot is Microsoft’s coding product, while OpenAI supplies models and is not a Microsoft-owned model developer.
PYMNTS also says a Microsoft spokesperson confirmed that the company has been steering staff toward GitHub Copilot, while engineers can still choose other models. That supports describing the change as an internal preference or procurement shift, rather than a blanket Claude ban.
The response also limits what can be inferred about Microsoft’s customer-facing products. Encouraging employees to use a preferred tool does not, by itself, establish that Microsoft is removing Anthropic models from products it sells.
Meta’s Reported Claude Code User Count Falls by Half
At Meta, PYMNTS reports that the number of employees using Claude Code fell from approximately 60,000 earlier in 2026 to around 30,000. The company reportedly encouraged employees to use Muse Code, described in the coverage as a Claude Code competitor, and its internal-only MetaCode tool.
That is a roughly 50% decline in reported users, but it does not necessarily mean Meta canceled 30,000 paid seats. The accounts describe employee usage without a verified licensing inventory. They do not establish whether the figures measure active users over a particular period, provisioned accounts, or another internal definition. Converting them directly into subscription losses would add precision the evidence does not support.
The cause remains partly uncertain. The Decoder attributes part of the decline to layoffs and part to Meta’s push toward its own tools; PYMNTS emphasizes the tool migration. Both accounts trace back to The Information, so they are not independent confirmations of the underlying figures.
The reported spending gives management reason to scrutinize usage. The Decoder says Meta spent more than $105 million on Claude Code during one 28-day period. That snapshot is not a verified recurring monthly bill and should not be annualized as though consumption stayed constant.
Separately, PYMNTS says Meta had been on track to spend billions of dollars on internal AI tools overall, with Claude Code the most popular. That broader budget should not be confused with spending exclusively on Anthropic.
Fewer users and lower spending are not interchangeable. A smaller group running intensive coding agents could still consume substantial resources. Conversely, changing how employees use a tool could reduce spending without removing their access.
PYMNTS says Meta did not comment in The Information’s report and did not immediately respond to its own request for comment. The cited coverage does not establish an Anthropic response.
Choosing Copilot Does Not Necessarily Mean Dropping Claude
Companies have straightforward strategic incentives to favor their own tools, though the reports do not establish every motive behind these decisions.
Using an internally preferred product can give a company more control over budgets, access policies, product integration, and development priorities. Employee usage can also provide practical feedback for improving that product. This practice, commonly called dogfooding, turns internal work into a source of product learning.
For Microsoft, directing engineering work toward GitHub Copilot can strengthen the coding interface it owns. Meta’s encouragement of Muse Code and MetaCode could help those tools mature while reducing reliance on an outside supplier. These are plausible strategic benefits, not confirmed explanations from the companies.
The interface employees use and the model doing the work, however, are separate decisions.
GitHub’s supported-model documentation lists models from Anthropic alongside OpenAI and other providers. Availability depends on the Copilot plan and where the product is used, and model availability can change.
An employee can therefore move from a standalone Anthropic tool into GitHub Copilot while continuing to use an Anthropic model, where that model is available and permitted. A migration into Copilot shows a change in product preference; it does not automatically mean all associated Claude consumption disappears.
Organization and enterprise owners can also enable or restrict specific models, according to GitHub’s documentation. Choosing Copilot as the approved coding environment and deciding which models employees may access within it are separate administrative choices.
For Anthropic, those outcomes have different consequences. Losing a direct Claude Code user may weaken its position as the employee’s primary coding tool. Continued model usage through another platform could preserve some consumption, although these reports do not establish the commercial terms or revenue impact.
Frequently Asked Questions
3 questions
1Are Meta and Microsoft banning Claude?
The reporting does not establish a blanket Claude ban at either company. PYMNTS says Microsoft is encouraging GitHub Copilot while allowing engineers to choose other models. Meta reportedly pushed employees toward Muse Code and MetaCode, but the available accounts do not establish a company-wide prohibition on Anthropic tools.
2
Sources
- The Information’s October 5, 2026 reporttheinformation.com
- PYMNTS’ accountpymnts.com
- The Decoder’s coveragethe-decoder.com
- supported-model documentationdocs.github.com
