Anthropic unintentionally revealed the details of an upcoming release, an exclusive CEO event, and other internal information, including images and PDFs in what seems to be a significant lapse in security.
Leaked Anthropic Information
The leaked information was made available by Anthropic’s content management system (CMS), which is sued to publish information to parts of the company’s website.
Close to 3,000 assets were linked to the website blog that were not previously published to the public-facing news or research sites that were publicly accessible in the data cache, according to Alexandre Pauwels, a cyber security researcher at the University of Cambridge Fortune asked to review the material.
Fortune informed the AI company of the issue on Thursday, March 26, 2026, and Anthropic took the steps to secure the information. But before the AI company took those steps, it stored the content on the website in a central system accessible without a login. Anyone with technical knowledge could make requests to the public-facing system to gain access to the files.
Even though some of this data had not been published to the company website, the underlying system would return the digital assets being stored to anyone who knew how to ask, including unpublished materials.
Most of the information appeared to be discarded or unused assets for blog posts, but some of the data appeared to contain sensitive, detailed information about upcoming products, including an unreleased AI model that Anthropic said represented a “step change” in AI capabilities, with significant improvements in “reasoning, coding, and cybersecurity.”
Cause of the Leak
It appears the issue stems from how the CMS works. All assets uploaded to the central data store are made public by default, unless specifically set to private. Anthropic did not restrict access to these documents that were not supposed to be made available to the public, resulting in the large cache of files being available to the public data lake, according to cybersecurity professionals who spoke with Fortune.
“An issue with one of our external CMS tools led to draft content being accessible,” an Anthropic spokesperson told Fortune who attributed the issue to “human error in the CMS configuration.”
AI-Generated Code SNAFUs
Lately, multiple tech companies have experienced technical problems due to AI-generated code or with AI agents. Anthropic, makes the Claude AI models and has boasted of automating a good portion of its internal software development using Claude-based AI coding agents, said AI was not the problem in this case.
The CMS issues was “unrelated to Claude, Cowork, or any Anthropic AI tools,” according to the spokesperson.
The AI tech company sought to downplay the significance of some of the data that was left unsecured. “These materials were early drafts of content considered for publication and did not involve our core infrastructure, AI systems, customer data, or security architecture,” according to the spokesperson.
This is not the first time a tech company has unintentionally exposed internal assets by leaving them publicly accessible before official announcements.
Twice, Apple leaked information through its website. Once, in 2018, upcoming iPhone names were publicly accessible hours before launch, and in 2025, a developer found that Apple shipped its redesigned App Store with debugging files still active, making the entire internal code of the site readable to anyone.
Epic Games and Nintendo have also leaked pre-released images, in-game assets, and other media via content delivery network (CDNs) or staging servers, similar to the the data lake in the Anthropic case. Even Google has accidentally exposed internal documents at public URLs and data associated with Tesla vehicles was exposed through misconfigured third-party servers.
Sources:
Fortune: Exclusive: Anthropic left details of an unreleased model, an upcoming exclusive CEO event, in a public database
CNBC: Cybersecurity stocks fall on report Anthropic is testing a powerful new model
Yahoo Finance: Cybersecurity Stocks Take a HitL Palo ALto Drops 6%, Okta Tumbles 7% on AI Competition Fears
Featured Image Courtesy of TORLEY’s Flickr Page – Creative Commons License
Discover more from Guardian Liberty Voice
Subscribe to get the latest posts sent to your email.