Anthropic's independent research body, The Anthropic Institute (TAI), has published its research agenda covering economic diffusion, threats and resilience, AI systems in the wild, and AI-driven R&D—including the risk of recursive AI self-improvement by 2028.
Anthropic launched Dreaming—a scheduled process that reviews agent session history and updates memory between tasks—as a research preview in Claude Managed Agents at Code with Claude 2026. Outcomes, multiagent orchestration, and webhooks also moved to public beta.
xAI is providing Anthropic with full access to Colossus 1, an AI supercomputer with 220,000+ NVIDIA GPUs. The deal immediately doubles Claude Code rate limits across all paid plans and removes peak-hour throttling for Pro and Max subscribers.
The Center for AI Standards and Innovation (CASI) secured agreements with Google DeepMind, Microsoft, and xAI to review frontier AI models for national security risks before launch. The policy shift follows alarm over Anthropic's Claude Mythos autonomous cybersecurity capabilities.
Cloudflare is laying off more than 1,100 employees globally, framing the restructuring not as cost-cutting but as a necessary redesign for the agentic AI era, with internal AI usage up 600% in three months.
A well-argued HN post explains why reliable AI agents require deterministic software scaffolding rather than ever-more-elaborate prompting, earning 552 upvotes from developers.
Google DeepMind has shared the progress of AlphaEvolve, its Gemini-powered coding agent, which has spent the past year discovering and improving algorithms across quantum computing, biotechnology, logistics, and Google's own AI infrastructure.
OpenAI has launched GPT-Realtime-2 in its API, bringing GPT-5-class reasoning to real-time voice interactions. The release also includes GPT-Realtime-Translate for live multilingual speech translation and GPT-Realtime-Whisper for streaming transcription.
A new arXiv paper (2605.00842) explains why fine-tuning an LLM on a narrow, harmless task can induce broad misalignment in unrelated contexts — attributing it to feature superposition geometry. The work gives a theoretical foundation to one of AI safety's most troubling open problems.
Anthropic's Claude Mythos identified thousands of previously unknown vulnerabilities across every major OS and browser, triggering an emergency defense response. The model is restricted to ~40 vetted U.S. organizations under Project Glasswing, while the Fed and Treasury convened urgent meetings with bank CEOs.
IBM CEO Arvind Krishna unveiled a comprehensive multi-agent AI operating model at Think 2026, simultaneously launching watsonx Orchestrate (next-gen), IBM Bob, IBM Concert, and IBM Sovereign Core to bridge the AI divide separating investment from ROI.
Apple is planning an "Extensions" feature for iOS 27 that lets users choose a preferred third-party AI — including Google Gemini and Anthropic Claude — to power Siri, Writing Tools, and Image Playground, marking the company's most open AI ecosystem move yet.