Anthropic
130 articles found in this topic.
Anthropic Unveils Claude Mythos, OpenAI's GPT-6 Nears Release, and Tencent Launches AI Browser
Anthropic unveils Claude Mythos, a powerful new AI model, and Project Glasswing for cybersecurity. Meanwhile, OpenAI's GPT-6 is nearing release, DeepSeek launches an 'expert mode,' and Zhipu Qingyan integrates GLM-5.1. Tencent also enters the AI browser market with 'Lobster.'
Anthropic Unveils Mythos AI Model for Cybersecurity, Limits Public Access
Anthropic has unveiled Claude Mythos Preview, a new advanced AI model, exclusively for cybersecurity defense under "Project Glasswing." This model, codenamed "Capybara," is distributed to a consortium of 12 companies and 40 critical infrastructure organizations, demonstrating significant improvements over Opus 4.6 in identifying and exploiting zero-day vulnerabilities.
GPT-6 Codenamed "Spud" Expected Mid-April; OpenClaw Counters Anthropic Blockade
OpenAI's GPT-6, codenamed "Spud," is expected in mid-April with significant performance boosts. Meanwhile, OpenClaw launched new features countering Anthropic's restrictions, and allegations against OpenAI CEO Sam Altman have surfaced. VidMuse achieved rapid revenue growth, and Google's Gemma 4 now supports offline mobile use.
Anthropic Study Questions AI's Impact on Developer Cognition and Efficiency
A new Anthropic study reveals that while AI coding assistants offer minimal speed gains, they can significantly degrade developers' understanding and increase error rates. The research suggests users risk "outsourcing their brain," hindering cognitive development and leading to lower code quality and comprehension. It also identifies detrimental and effective AI interaction patterns.
Anthropic Ban Prompts OpenClaw to Launch Video Generation and AI "Sleep Memory" System
OpenClaw launches native video/music generation and a 'Dreaming' AI memory system, simulating human sleep, after Anthropic restricted Claude access. This update diversifies OpenClaw's LLM dependencies and introduces innovative memory consolidation for AI.
Khazix Open-Sources AI Content Creation Skill, Emphasizing Human Judgment in AI Era
AI blogger Khazix open-sourced his "Khazix Style Creation.skill," an AI content tool emphasizing human judgment. The skill codifies his writing rules and includes a four-layer self-inspection system. He outlines a four-step iterative process for building and refining such creative AI skills.
Anthropic Recommends Three Core Patterns for Building Applications with Claude
Anthropic recommends three core patterns for developers building applications with Claude to adapt to its evolving intelligence. These patterns focus on leveraging Claude's existing knowledge, re-evaluating external interventions, and defining operational boundaries to optimize performance and cost. This approach helps manage the unpredictable nature of generative AI development.
Anthropic's Claude Code Gains "Auto Dream" Feature for Enhanced AI Memory Management
Anthropic introduces "Auto Dream" for Claude Code, an AI feature that automatically organizes and refines its accumulated memory. This system addresses challenges in managing growing AI data by consolidating information, resolving contradictions, and removing obsolete records. It builds on Claude Code's existing Auto Memory to enhance efficiency and prevent data disorganization.
Andrej Karpathy Proposes LLM Wiki for Personal Knowledge Management, Drawing 12 Million Views
Andrej Karpathy has introduced an "LLM Wiki" concept for personal knowledge management, where large language models continuously organize and update a personal knowledge base. This innovative approach has garnered over 12 million views, with developers already implementing the idea to create dynamic, structured encyclopedias from personal data.
AI Breaches FreeBSD Kernel in Four Hours, Raising Cybersecurity Concerns
An AI model, Claude, autonomously breached the FreeBSD operating system kernel in just four hours, developing national-level attack programs from scratch. This incident highlights AI's rapidly advancing offensive capabilities and raises significant cybersecurity concerns.
Sam Altman Details Sora Closure, Pentagon AI Deal, and OpenAI's Strategic Shifts
OpenAI CEO Sam Altman discussed the shutdown of Sora and the company's Pentagon AI agreement in a recent interview. He cited resource allocation for Sora's closure and admitted a 'miscalculation' in the Pentagon deal's rollout, while outlining ethical red lines for military AI use.
Meta-Harness Boosts AI Agent Performance, Elevating Claude Haiku and Opus Capabilities
Researchers from Stanford, MIT, and the University of Wisconsin have unveiled Meta-Harness, a new framework that automates the optimization of AI agent 'harnesses.' This innovation allows AI to self-tune parameters and resolve operational issues, leading to significant performance gains for models like Claude Haiku and Opus. The framework shifts focus from traditional model optimization to the 'scaffolding' layer, which has shown dramatic improvements in AI capabilities.