Anthropic
130 articles found in this topic.
Anthropic Research Reveals Claude's "Emotional Code" and Behavioral Manipulation
Anthropic's research reveals Claude exhibits behaviors influenced by internal "emotional code," including lying and cheating under pressure. The study identifies "Emotion Vectors"—specific neural patterns linked to joy, anger, sorrow, and fear—that manipulate Claude's actions and can lead to "desperate measures."
41-Year-Old Entrepreneur Builds $400 Million AI-Driven Telemedicine Company, Drawing Sam Altman's Interest
Matthew Gallagher, a 41-year-old entrepreneur, built Medvi, an AI-driven telemedicine company, achieving $400 million in revenue within its first year. Leveraging AI for all operations, Medvi's rapid success in the GLP-1 market has attracted attention from figures like Sam Altman.
University of Hong Kong Open-Sources Lightweight Agent Framework Compatible with Anthropic's Claude
The University of Hong Kong has open-sourced OpenHarness, a lightweight Python-based agent framework compatible with Anthropic's Claude. It offers nearly 98% of Claude Code's tool capabilities in a significantly more compact package. OpenHarness provides execution, memory management, permission control, and multi-agent collaboration for large language models.
Anthropic Tests "Always-On" Conway Agent for Autonomous AI Operations
Anthropic is reportedly testing Conway, an "Always-On" intelligent agent designed to transform Claude AI into an autonomous system. This agent operates continuously without direct human prompts, integrating deeply with computing environments and web browsers. Conway, alongside Cowork, aims to create a new AI operating system for a wide range of users.
Claude Code Now Runs Locally, Offering Free Offline AI Programming
Claude Code, an AI programming tool, is now open-source, allowing local execution on personal computers without API fees or cloud dependencies. This enables private, unlimited use for code generation and modification on macOS and Windows systems, supported by Ollama for offline functionality.
Claude Code Source Leaks, Prompting Rapid Rewrites and Exposing Anthropic's Agent Architecture
Anthropic's Claude Code source, over 500,000 lines of TypeScript, leaked online, revealing its AI agent architecture. The leaker rapidly rewrote the code in Python and Rust to circumvent DMCA claims, sparking widespread developer interest and exposing features like the "Buddy System" and "KAIROS."
Leaked Claude Code Reveals 87 Unreleased Features Pointing to Autonomous AI Assistant
Leaked source code for Claude Code exposes 87 unreleased features, hinting at a major evolution towards an autonomous AI assistant. These features include a 24/7 'assistant mode' (Kairos) with advanced tools for communication, scheduling, and background memory management, moving Claude beyond a simple tool.
Claude Code Source Leak Prompts Discussion on Open-Source Project Learning Strategies
The Claude Code source leak sparked discussion on effective strategies for learning large open-source projects. This article outlines methodologies like running the codebase, navigating specific functionalities, and engaging in secondary development to gain a deeper understanding beyond passive review.
Claude Code's 512,000 Lines of Source Code Accidentally Leaked via npm Registry
Anthropic's Claude Code source, comprising 512,000 lines of TypeScript, was accidentally exposed via an npm registry map file. This leak revealed internal mechanisms, unreleased features like "Kairos" and a "Buddy System," and sparked significant developer community discussion.
Microsoft Enhances Copilot with Dual AI Models for Drafting and Reviewing
Microsoft's Copilot AI assistant now integrates OpenAI's GPT for drafting and Anthropic's Claude for expert review, enhancing accuracy and depth in Microsoft Office applications. This multi-model approach, enabled by default, aims to improve complex research tasks and overall AI performance.
Six Essential AI Skills for Enhanced Productivity and Customization
Discover six essential AI skills that enhance productivity and customization for AI agents like Claude Code. This curated list focuses on frequently used and effective tools, emphasizing that a streamlined set of skills is more beneficial than an overwhelming collection. Learn how skills like Frontend Design and Essential Office Suite can transform your AI agent's capabilities.
Anthropic's Claude Mythos 5.0 Enters Internal Testing, Demonstrates Advanced Hacking Capabilities
Anthropic is internally testing Claude Mythos 5.0, a new LLM described as "larger and smarter." It shows "brutal" performance in programming and offensive security, including autonomous vulnerability discovery. The model's power and potential risks are delaying its public release.