The leaked code, exceeding 512,000 lines, seems to reveal unannounced functionalities, guidelines for Claude, and additional insights.
The leaked code, exceeding 512,000 lines, seems to reveal unannounced functionalities, guidelines for Claude, and additional insights.
Following the release of Anthropic’s Claude Code’s 2.1.88 update, individuals rapidly noticed that it housed a package with a source map file showcasing its TypeScript foundation, with one user on X highlighting the leak and sharing a document containing the code. The disclosed information is said to encompass more than 512,000 lines and offers a glimpse into the functioning of the AI-driven coding tool, as earlier reported by Ars Technica and VentureBeat.
Individuals who examined the code assert that they have unveiled upcoming functionalities, directives from Anthropic regarding the AI bot, and details about its “memory” structure. Among the findings reported by individuals is a Tamagotchi-like companion that “sits beside your input box and responds to your programming,” according to a Reddit post, in addition to a “KAIROS” function that may allow for a constant background agent. Users also encountered a remark from one of Anthropic’s developers, who confesses at one point that the “memoization here increases complexity significantly, and I’m not convinced it actually enhances performance.”
Although Anthropic subsequently resolved the issue, it did not deter users from duplicating the code to a GitHub repository, which has now accumulated over 50,000 forks (or replicas of the repository). Anthropic introduced Claude Code in February 2025, and the tool gained traction following the addition of agentic capabilities that execute tasks on behalf of the user.
“Earlier today, a Claude Code release featured some internal source code. No private user data or credentials were at risk or revealed,” Anthropic representative Christopher Nulty mentions in an emailed statement to The Verge. “This was a packaging error due to human mistake, not a security lapse. We’re implementing measures to ensure this does not reoccur.”
Arun Chandrasekaran, an AI analyst at Gartner, informs The Verge that while the leak of Claude Code presents “risks including providing malicious actors with potential means to circumvent precautions,” its long-term effect might be confined to motivating Anthropic to enhance its tools and processes for improved operational maturity.
