Meta and Microsoft are reducing employee use of Anthropic’s Claude AI as they prioritize internal coding tools and tighter spending controls. The changes concern employee workflows and budgets, rather than ending customer access to Claude through Microsoft’s products.
The reported pullback highlights a conflict: both companies buy Anthropic technology while competing with its coding assistants. Developer productivity, infrastructure costs, and control over software engineering workflows are shaping those purchasing decisions.
Microsoft expected internal spending on Anthropic technology to reach at least $1 billion annually. That projection reportedly declined by more than a third after executives encouraged employees to reduce Claude usage and favor Microsoft tools, including GitHub Copilot and OpenAI models.
Meta and Microsoft Scale Back Claude AI
Within Microsoft’s cloud and AI division, monthly AI spending caps reportedly fell from $100,000 per employee to $10,000 in most cases. These amounts represent authorized limits, not spending by every employee. Some engineers expressed frustration over reduced flexibility to experiment with models.
The distinction between internal consumption and customer availability remains. Customer spending on Anthropic models through Microsoft’s enterprise platforms reportedly continues growing despite restrictions on employees’ usage.
At Meta, Claude Code users reportedly dropped from 60,000 earlier this year to around 30,000. Layoffs contributed, but the report attributed the larger impact to efforts promoting Meta’s assistants, MetaCode and Muse Code.
MetaCode reportedly exceeded 30,000 internal users, while Muse Code attracted more than 6,000 employees. Meta began testing Muse Code with external customers in August.
According to CSN, Meta reportedly spent more than $105 million on Claude Code during one 28-day period, indicating that declining user counts do not automatically translate into modest operating costs.
Switching coding assistants does not eliminate risks associated with automated command execution, repository access, and exposed credentials.
Disclosed Claude Code vulnerabilities, CVE-2025-59536 and CVE-2026-21852, involved repository configuration settings that could bypass consent controls or redirect authenticated API traffic.
Attackers could exploit these mechanisms for unauthorized execution and API key theft. Anthropic implemented fixes before public disclosure.
Microsoft’s ecosystem has faced comparable exposure. The patched RoguePilot vulnerability exploited malicious instructions embedded in a GitHub Issue to influence Copilot inside Codespaces.
Its attack chain combined terminal commands, symbolic links, and remote schema requests to steal a repository token and enable takeover.
For security teams, these incidents reinforce the importance of restricted agent permissions, trusted project configurations, and scrutiny of automated actions.
Budget reductions and vendor consolidation address commercial priorities; they do not replace controls around credentials, untrusted repository content, or execution boundaries. The shift reflects competitive pressure, not a complete departure from Claude.
Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team
