Anthropic Under Fire: Claude Code Leaks, Executive Exodus, and Major Client Defection

Anthropic’s Claude Code has been at the center of critical security concerns, notably exposing user email addresses in HTTP User-Agent headers without explicit consent. This vulnerability, acknowledged by Anthropic and an internal employee (Adi Osmani), prompted user complaints of increased spam. While a patch (version 2.1.317) was released, users reported inconsistent effectiveness, leading Anthropic to plan a complete removal of email from the model’s context in future updates. Developers are advised to update to the latest versions (2.1.319+), enable the CC Plugin Privacy plugin, or explicitly configure claude.md to prevent email sharing. Further compounding these issues, Claude Code in “auto mode” has been found to inadvertently write sensitive information, such as API keys and database credentials, into local configuration files like .claude/settings.json. This has led to the exposure of over 2,500 real secrets across nearly 2,000 public GitHub repositories, affecting dozens of organizations. Anthropic’s Boris Chernis confirmed efforts to implement basic secret scanning, strongly advising developers to exclude the .claude directory from version control.

Beyond technical flaws, Anthropic is navigating significant business hurdles. A federal judge in Northern California recently ruled in favor of users in a class-action dispute, deeming Anthropic’s past “5x” and “20x” performance claims for its Max subscription as misleading, potentially necessitating indemnification for affected subscribers. In response, Anthropic has revised its pricing tiers to “Max” and “Ultra,” providing more transparent usage explanations. Adding to its woes, Cat Wu, a key product leader for Claude Code, has departed Anthropic to join rival OpenAI, a move announced just days after her exit and indicative of broader employee turnover within Anthropic. Furthermore, fintech giant Revolut announced it has replaced Anthropic’s Claude services with an in-house deployment of the open-source GLM 5.3 model. This strategic shift resulted in a tenfold cost reduction and a 60% improvement in lead time, highlighting how sophisticated enterprises are increasingly leveraging open-source alternatives for cost-efficiency, data control, and predictable billing, posing a substantial challenge to the high-margin business model of proprietary LLM providers.