Meta and Microsoft have reportedly restricted their employees' access to Anthropic's AI tool Claude, while promoting their own AI programming tools. These changes mainly affect how internal teams use AI for coding and other tasks, rather than limiting customer access to Claude through Microsoft's products. The move underscores increasing pressure to manage costs as AI coding assistants become more integrated into daily software development processes. Microsoft had canceled most internal licenses for Claude Code, a coding tool developed by Anthropic, and began steering its developers toward GitHub Copilot CLI, a tool Microsoft itself has developed. This decision came nearly five months after Microsoft expanded access to Claude Code for thousands of employees, including engineers, project managers, and designers. The move appears to be both strategic and financially motivated. In July, Microsoft began shifting work within its own products, reducing reliance on partners like OpenAI and Anthropic in favor of internally developed AI models. Microsoft CEO Satya Nadella highlighted that these internal models have already outperformed leading general-purpose models in many areas while using far fewer computational resources, measured in "tokens." Tools like GitHub Copilot, Excel, and Outlook have already shown promising results, with Copilot Chat and PowerPoint next in line for integration. At Meta, the number of internal users of Claude Code reportedly dropped from about 60,000 at the start of the year to around 30,000. While layoffs contributed to this decline, the main reason appears to be Meta's push for its own AI tools, such as MetaCode and Muse Code, which use Meta's internal models. MetaCode had more than 30,000 internal users, while Muse Code had over 6,000. Meta began testing Muse Code with external customers in August. Despite the shift, Meta reportedly spent more than $105 million on Claude Code over a 28-day period, showing that reduced user numbers don’t always mean reduced spending. Microsoft had previously expected to spend at least $1 billion annually on Anthropic’s technology. However, after executives asked employees to use Microsoft tools more frequently, including GitHub Copilot and OpenAI models, this forecast dropped by more than a third. Within Microsoft’s cloud and AI division, monthly AI spending caps for employees were reportedly reduced from $100,000 to about $10,000 in most cases. These caps apply to spending limits, not the actual amounts spent by employees, and have frustrated some engineers who previously had more freedom to test different AI models. Despite these internal restrictions, customer spending on Anthropic’s models through Microsoft’s enterprise platforms continues to grow. Some reports have identified security issues in Claude Code related to repository settings, which could allow unauthorized execution or API key theft. Anthropic addressed these issues before they were made public. In February, Amazon also imposed restrictions on Claude Code, requiring formal approval for its use in coding tasks. Amazon encourages employees to use its internal AI tool, Kiro, instead of third-party solutions like Claude Code, a move that has drawn some internal criticism. Employees have questioned why an unofficially approved internal tool is promoted to customers. However, Amazon has not imposed a complete ban, instead emphasizing stricter internal standards for production tools. Microsoft’s tools have faced similar security concerns. A vulnerability called RoguePilot, now fixed, showed how malicious instructions in a GitHub ticket could lead to the takeover of a repository. These findings highlight the importance of limiting permissions, ensuring reliable project settings, and carefully reviewing AI actions. Despite these challenges, Anthropic’s annualized revenue rate of $65 billion shows that demand for its models remains strong. The reduction in internal use by companies like Microsoft and Meta reflects tighter budgets and increased competition, rather than a complete loss of interest in Claude.