In a recent software update, a debug file with almost 500,000 lines of internal code was accidentally exposed. Developers around the globe duplicated the leaked code to grasp how Anthropic utilizes the large-scale language model (LLM) powering the application and uploaded it on the coding platform GitHub.
A spokesperson for Anthropic verified the occurrence. They stated that this was an "issue related to a human error, not a security breach".
Anthropic requested that GitHub take down over 8,000 versions and adaptations of Claude Code's source code that developers posted on the platform. The company subsequently refined its criteria to 96 protected repositories.
An Anthropic representative stated that the event did not allow users access retrieving customer information or the weights (internal calculations) of AI models.
Nonetheless, the leak exposed Anthropic's techniques, resources, and guidance that enable users to manage and steer these models, according to the WSJ. Consequently, Anthropic's rivals and many startups and developers now possess a comprehensive plan for replicating Claude Code's functionalities. Hackers acquired details to identify vulnerabilities they could exploit to infiltrate Claude Code's software or manipulate Claude's AI model for cyberattacks, the newspaper reports.
source: wsj.com
A spokesperson for Anthropic verified the occurrence. They stated that this was an "issue related to a human error, not a security breach".
Anthropic requested that GitHub take down over 8,000 versions and adaptations of Claude Code's source code that developers posted on the platform. The company subsequently refined its criteria to 96 protected repositories.
An Anthropic representative stated that the event did not allow users access retrieving customer information or the weights (internal calculations) of AI models.
Nonetheless, the leak exposed Anthropic's techniques, resources, and guidance that enable users to manage and steer these models, according to the WSJ. Consequently, Anthropic's rivals and many startups and developers now possess a comprehensive plan for replicating Claude Code's functionalities. Hackers acquired details to identify vulnerabilities they could exploit to infiltrate Claude Code's software or manipulate Claude's AI model for cyberattacks, the newspaper reports.
source: wsj.com