Official announcements from AI vendors

Anthropic Enhances Transparency by Publishing System Prompts for Claude AI Models

Aug 26, 2024, 07:56 PM

Anthropic, the company behind the AI model Claude, has taken a significant step towards transparency by publishing the system prompts that guide its AI models' behavior. These prompts, which act as guidelines for the AI's responses, have been made available for all three versions of Claude, including Sonnet, Opus, and Haiku. The release includes a changelog, marking the first instance of a major AI vendor providing such detailed documentation. The latest prompts, dated July 12, specify what the Claude models cannot do, such as opening URLs, links, or videos, and avoiding facial recognition. This move is seen as a positive step towards industry-wide transparency. The system prompt changes are logged for Claude's mobile apps, but do not affect the API.