Claude’s Code Leak Exposes Unreleased Features from Anthropic

Anthropic prepares for IPO; Claude AI developer likely to launch public offering in October.
Anthropic PBC’s unintentional release of source code for its widely used AI coding assistant was attributed to “process errors” stemming from the company’s accelerated product launch schedule, as stated by a senior executive.

Paul Smith, the chief commercial officer of Anthropic, emphasized that the leak was “absolutely not breaches or hacks,” assuring that the issues have been resolved. “They’re part of the incredibly rapid release cycle that we’ve had around Claude Code,” Smith mentioned during an interview on Wednesday.

Anthropic approaches the situation “incredibly seriously,” he added, pointing out that the firm has “all the right people focused on addressing it.”
This accidental release represented Anthropic’s second security incident in just a few days, putting approximately 1,900 files and 512,000 lines of code connected to Claude Code at risk. Last week, Fortune reported that Anthropic had been unintentionally hosting thousands of internal documents on a publicly accessible server, which included a draft blog post discussing an upcoming model referred to internally as both “Mythos” and “Capybara.”

In a series of posts overnight on X, Claude Code creator Boris Cherny explained that Anthropic’s “deploy process has a few manual steps, and we didn’t execute one of the steps correctly.” He mentioned that the company has already “made a few improvements to the automation for the next time,” with additional enhancements planned.

The leaked code contained at least eight unreleased features, as noted by entrepreneur Abhishek Tiwari, who quickly developed a website to monitor the leak using AI. Among these features, he highlighted:

Kairos, a setting that enables Claude to operate in the background.

Coordinator Mode, which allows the AI to divide a task into components and assign them to individual workers.

Auto-Dream, where Claude reviews what it learns and organizes notes into well-structured memory files.

Ultraplan, allowing the creation of a separate cloud instance that can investigate and plan for up to 30 minutes at a time.

Anthropic did not provide comments regarding the features or future plans. “We’re always experimenting with new ideas. 90% don’t ship because we don’t think they’re good enough experiences,” Cherny replied to a post on X.

 

Also Read: Trump administration gears up for new tariffs on select drugmakers

Previous Article

Tax Breaks on Overseas Remittances: Implications for Travelers and Students

Next Article

11:11 | Trump Intensifies Iran Conflict Stance, Gold Prices Fall, UPI Achieves Record Transactions, Artemis II Launch, and Additional Updates