Claude Leak Shows That Anthropic Is Tracking Users’ Vulgar Language and Deems Them “Negative”

Anthropic, the AI company behind the popular chatbot Claude, has found itself in the spotlight after a significant leak of its source code. The leak, which occurred earlier this week, has triggered a flurry of activity as Anthropic representatives scramble to remove copies of the stolen work from the internet. The exposed code has allowed developers to reverse engineer aspects of Claude, raising concerns that competitors could gain a significant advantage. However, the leak has also revealed intriguing details about Anthropic's practices and future plans.

One of the most surprising revelations from the leaked code is that Anthropic is actively tracking users' use of vulgar language. Developers have discovered that Claude has a regular expression (regex) that detects phrases such as "wtf," "ffs," "piece of s***," "f*** you," and "this sucks." This information is then logged in the system, with a flag indicating whether the user's input is deemed "negative." While this feature does not change the behavior of the AI, it does provide insights into how users interact with the platform.

The tracking of vulgar language use has sparked debate among developers and users alike. Some have questioned whether Anthropic is using this data to gauge user satisfaction or to improve the AI's ability to understand and respond to negative emotions. In response to these concerns, Boris Cherny, the creator of Claude, clarified that the "f***s" chart is one of the signals they use to determine if people are having a good experience. He explained that the data is displayed on a dashboard to monitor user interactions and provide valuable feedback for improvement.

In addition to the language tracking, the leaked code has also shed light on Anthropic's internal processes. Developers have found evidence of a mood classification system that is exclusive to employees. When an Anthropic employee uses language that is deemed negative, a prompt appears asking them to share their transcript, effectively encouraging them to report any issues they may be experiencing. This feature raises questions about how Anthropic manages its team's well-being and handles feedback from its workforce.

The leak has also revealed plans for future features, such as an experimental "buddy" system that resembles a Tamagotchi. This feature is described as sitting beside the input box and reacting to the user's coding. While the exact purpose of this system is unclear, it suggests that Anthropic is exploring new ways to engage users and enhance their interaction with AI.

The Anthropic leak has provided a unique glimpse into the inner workings of a major AI company. It has not only highlighted concerns about data tracking and user privacy but also revealed intriguing insights into the company's development practices and future plans. As Anthropic works to mitigate the impact of the leak, the broader AI community is left to ponder the implications of such transparency and the potential for competitors to leverage this information to their advantage.