📰 News
AI News — page 6
Funding, launches, analysis, and interviews.
Musk v OpenAI trial enters week two as Brockman testifies on control battle
Greg Brockman testified that Elon Musk demanded majority control of OpenAI's for-profit arm in 2017, contradicting Musk's claims about preserving the nonprofit mission.
InstantDB launches GETadb.com for AI agents to build full-stack apps
InstantDB's new GETadb.com service provides AI coding agents with instant backend credentials and database access without requiring human sign-up or configuration.
Domain-Trained Legal AI Model Beats Frontier LLMs at 97% Lower Cost
Olava Extract, a specialized legal AI model, outperformed five frontier large language models on contract extraction tasks while reducing inference costs by up to 97%.
Researchers Develop Lightweight Method to Detect AI Hallucinations in Black-Box Models
New Distribution-Aligned Adversarial Distillation technique uses tiny proxy models to estimate uncertainty in commercial LLMs without accessing internal parameters or expensive sampling methods.
Researchers Propose BADIT Method to Reduce Cross-Task Interference in LLM Training
New research introduces Basic Abilities Decomposition for multi-task Instruct-Tuning (BADIT), which decomposes large language model parameters into orthogonal components to mitigate conflicting gradients during multi-task training.
Researchers Develop TurnGate Defense Against Multi-Turn AI Attacks
Academic researchers created TurnGate, a monitoring system that detects when multi-turn conversations with AI models reach the point where harmful responses could enable malicious actions.
OpenAI Codex Expands Beyond Coding as Claude Adds Creative Tools
OpenAI repositioned Codex for general knowledge work while Anthropic launched Claude Security and added support for creative applications like Blender and Adobe Creative Cloud.
CyberSecQwen-4B Matches 8B Cybersecurity Model at Half the Size
Researchers fine-tuned a 4B parameter model that achieves 97% of an 8B cybersecurity model's accuracy while running locally on consumer hardware.
Anthropic announces Code with Claude developer conference
Anthropic will host its first developer conference featuring hands-on workshops, live demos of new Claude capabilities, and conversations with the engineering teams.
Anthropic publishes engineering insights on building reliable AI agents
Anthropic has released a comprehensive collection of engineering blog posts detailing how the company builds and evaluates AI agents, covering everything from sandboxing to multi-agent systems.
Anthropic reveals natural language autoencoders to decode Claude's internal thoughts
Anthropic's interpretability team developed a technique to translate Claude's numerical internal representations into human-readable text, advancing AI transparency research.
Anthropic launches Academy learning platform with Claude certification courses
Anthropic has launched Anthropic Academy, a comprehensive learning platform offering certification courses on AI fluency, API development, and Claude implementation for developers and enterprises.
Anthropic ships Claude Opus 4.7 with adaptive thinking and 1M context
Anthropic released Claude Opus 4.7, its most capable model featuring adaptive thinking that adjusts reasoning depth based on task complexity and a 1M token context window.
Anthropic launches Claude Code agentic coding system
Anthropic released Claude Code, an autonomous coding agent that reads codebases, makes multi-file changes, runs tests, and commits code without line-by-line guidance.
Anthropic launches Claude Cowork for autonomous desktop tasks
Anthropic released Claude Cowork, an agentic AI that works autonomously on desktop computers to handle multi-step knowledge work tasks from file organization to document preparation.
Anthropic ships Claude Sonnet 4.6 with 1M token context window
Anthropic released Claude Sonnet 4.6, a hybrid reasoning model featuring a 1M token context window and pricing starting at $3 per million input tokens.