Anthropic has confirmed that human error led to the leak of the source code for its AI agent, Claude Code, which wiped trillions from global stock markets. The company described the incident as a release error rather than a security breach. The AI startup revealed that a packaging issue unintentionally exposed part of its internal code.“No sensitive customer data or credentials were involved or exposed. This was a release packaging issue caused by human error, not a security breach. We’re rolling out measures to prevent this from happening again,” an Anthropic spokesperson said. The leak, which surfaced online via a publicly accessible npm package, reportedly exposed around 2,200 files and roughly 30MB of TypeScript code. The incident quickly gained attention, with a post linking to the code drawing millions of views, and raised concerns about how much insight it could give competitors into Anthropic’s development approach.
Another Anthropic data leak incident raises broader concerns
This disclosure comes on the heels of another incident where internal documents, including the company’s plans for an upcoming AI model, were discovered in a publicly accessible data cache. The incident has brought the company’s internal practices into the spotlight.The developers who looked into the leaked code found unreleased features, such as an always-on background agent, also known as Kairos, and a companion feature with gamification elements. The code also hinted at features for multiple AI agents and permission approvals.At the same time, the codebase’s structure drew mixed reactions, with reports highlighting large, complex files and workarounds for circular dependencies.Anthropic, founded in 2021 by former OpenAI researchers, has seen rapid growth with its Claude family of AI models. Claude Code, launched in May, is used by developers to write code, fix bugs, and automate workflows and has contributed to rising competition with companies such as OpenAI, Google, and xAI.The company is also said to be exploring a potential IPO later this year, adding pressure as it works to address repeated lapses.
