Exclusive: Anthropic left details of unreleased AI model, exclusive CEO event, in unsecured database

Key Points:

• Anthropic inadvertently exposed nearly 3,000 unpublished digital assets, including draft blog posts and sensitive documents, through its content management system (CMS) which was publicly accessible without login credentials.
• The exposure resulted from a CMS configuration error that left uploaded assets public by default unless manually restricted, a mistake Anthropic attributed to human error rather than AI tools like its Claude models.
• Among the leaked materials were details of an unreleased, highly capable AI model and information about an upcoming invite-only CEO retreat, although the company emphasized no core infrastructure, customer data, or security systems were compromised.
• After being notified by Fortune, Anthropic secured the data and stated the exposed content mainly consisted of early drafts and internal assets, downplaying the significance of the leak.
• This incident follows a pattern of tech companies unintentionally leaking pre-release or internal information due to misconfigured public-facing systems, a risk potentially heightened by the increasing use of AI coding tools.