Anthropic is having a month

Back-to-Back Setbacks for the Cautious AI Company

Anthropic has long promoted itself as a leader in responsible AI development, consistently sharing in-depth work about AI risks and championing the ethical responsibilities that come with building advanced technology. The company is so committed to this message that it's currently engaged in a legal tussle with the U.S. Department of Defense over AI-related concerns.

However, two recent incidents have tested Anthropic's reputation for caution. Last Thursday, the company mistakenly made nearly 3,000 internal documents public, which included confidential information on an unannounced AI model. This week, they faced a new blunder: while releasing version 2.1.88 of their Claude Code software, Anthropic inadvertently exposed almost 2,000 source code files, totaling over 512,000 lines and essentially revealing the architecture for one of their flagship tools.

Claude Code Leak: What Happened?

The oversight was quickly spotted by a security researcher who shared the findings online. Anthropic clarified that this was a "release packaging issue" caused by human error, not a cyberattack. Regardless, the incident laid bare the inner workings of Claude Code, a command-line interface that has attracted attention in the industry, even prompting competitive responses from rivals like OpenAI.

The leak did not include the AI model itself, but rather the infrastructure around it, detailing how the system operates and enforces boundaries. Developers swiftly began dissecting and analyzing the code, debating whether the exposure would have longer-term consequences or simply fade into the next wave of AI innovation.

For Anthropic, these repeated missteps are likely to spark internal reflection and reinforce just how quickly things can slip, even at companies built on trust and caution.

Read the original article by Connie Loizos on TechCrunch for further details.