Anthropic tightens release checks after Claude Code leak

Anthropic is tightening internal release processes after internal source code for Claude Code, its AI coding assistant, was accidentally made public and widely shared online.

GitHub notices show the company filed copyright takedown requests that removed thousands of copies. The takedowns affected more repositories than intended, and were scaled back.

Around 1,900 files and 512,000 lines of code were exposed. No customer data or credentials were involved, calling it a packaging error rather than a security breach.

Claude Code creator Boris Cherny said a missed manual step in the deploy process caused the release, and that Anthropic has added automation safeguards.

A security expert warned the code could act as a blueprint that helps attackers understand and potentially bypass protections.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Leak revealed unreleased features and business stakes

Coverage centered on the mistake, yet the larger impact sits with Claude Code’s revenue, with VentureBeat citing market data saying Claude Code alone reached $2.5 billion in annualized recurring revenue (ARR) 1.
Rivals can treat the released material as a playbook, since it outlines work beyond a basic AI layer on top of a coding assistant 1.
Details include a background agent codenamed KAIROS, which could support an always-on background agent, plus a three-layer memory architecture meant to improve efficiency 1, 2.
It also described “Undercover Mode,” which tells the tool not to reveal Anthropic-internal information in public/open-source repos, along with a Tamagotchi-like “pet” feature aimed at boosting engagement 1, 2, 3.

Technical slip meets a government standoff

The incident landed during Anthropic’s lawsuit against the U.S. Department of Defense over a “supply chain risk” designation or blacklist tied to the military supply chain 4, 3.
The government argues Anthropic could “surreptitiously alter the behavior of the model” in sensitive environments 3.
The Register says the material includes extensive data logging and retention in Claude Code, plus “UNDERCOVER” guidance for open-source repositories, which could strengthen the government’s position 3.
This adds pressure for partners such as Microsoft, which has up to $5 billion invested in Anthropic and filed an amicus brief (a “friend of the court” filing meant to influence a case) backing Anthropic while citing risk to Microsoft products including Azure and Copilot offerings that support Claude 4.