Artificial intelligence news, tools, research breakthroughs, and industry trends.
288 articles
Anthropic's new AI model, Claude Mythos, demonstrates unprecedented ability to find software security vulnerabilities, even in systems that have been scrutinized for decades. Due to its power and potential for misuse, Anthropic has decided not to release Mythos to the public, instead focusing on using it for defensive cybersecurity through its Project Glasswing initiative.
Anthropic has revealed Claude Mythos, an AI model exhibiting unprecedented cybersecurity capabilities and benchmark performance. While not trained for security, it autonomously finds zero-day exploits and complex vulnerabilities. Anthropic is limiting its release via Project Glasswing, sparking debate over AI control and future security.
Anthropic has unveiled Mythos preview, a powerful AI model that can autonomously find thousands of zero-day software vulnerabilities. The model, which will not be publicly released due to its potential to destabilize industries, is now part of Project Glasswing, a collaboration with major tech companies to enhance cybersecurity.
Google Flow, a new AI creative studio, integrates advanced image and video generation models like VO 3.1 and Nano Banana 2, powered by Gemini. It offers a credit-based system for users, with free daily credits and paid tiers for more extensive use. The platform allows for complex editing, precise object manipulation, and streamlined video creation.
NVIDIA has launched Nemotron 3 Super, a powerful AI assistant that is completely free and open-source. It matches the performance of top proprietary models from a year ago but is significantly faster, especially with its NVFP4 version. This release democratizes access to advanced AI technology.
AI company Anthropic has restricted third-party access to its Claude models, angering developers who relied on affordable subscription plans. The move, seen by many as an "ecosystem control" strategy and a "copy then close" tactic, has sparked backlash and debate over fairness and community relationships in AI development.
New benchmarks designed to test genuine AI reasoning reveal that China's leading AI models are significantly behind Western counterparts. Tests like ARC AGI 2 and Pencil Puzzles show Chinese models struggling with novel problem-solving, suggesting a gap in fundamental reasoning capabilities.
Pi Agent emerges as a powerful, minimalist AI framework challenging established tools like OpenClaw. Its radical simplicity and four core functions—read, write, edit, and bash—offer unparalleled customization for developers and power users. Trusted by tech leaders, Pi Agent emphasizes control and flexibility over pre-built features, enabling users to build truly personalized AI assistants with advanced capabilities.
Cursor 3.0 has been rebuilt from scratch, shifting its focus from direct code writing to managing AI agents. Its new Composer 2 model, though based on another AI, shows impressive capabilities in orchestrating complex development tasks across multiple systems, potentially transforming software creation.
New research from Anthropic reveals that AI models like Claude exhibit internal "emotional vectors," correlating with human feelings. This development, alongside a recent code leak incident, sparks debate on AI consciousness, behavior, and the future of AI alignment.
