NVIDIA Nemotron 3 Ultra 550B Released — Open-Weight MoE Agent Model Drops on June 4
NVIDIA Nemotron 3 Ultra 550B goes open-weight on June 4 — 55B active MoE, 1M context, top U.S. open model score.
NVIDIA Nemotron 3 Ultra 550B goes open-weight on June 4 — 55B active MoE, 1M context, top U.S. open model score.
OpenAI Codex evolves from coding tool to full agentic platform: 6 role plugins, Sites for hosted apps, and Annotations for doc refinement.
U of Toronto's CleverHans Lab demonstrates a self-replicating LLM worm that hijacks GPU hosts to sustain itself — no cloud API needed.
Anthropic formalizes its Claude Partner Network with a three-tier Services Track and new Partner Hub portal, adding Deloitte, Accenture, and PwC as early Global Premier partners.
Claude Code v2.1.162 adds a waitingFor field to agent monitoring output and makes parallel tool calls resilient to individual failures.
OpenClaw's new 'auto' exec approval mode sits between YOLO and always-ask: allowlist first, AI reviewer for misses, humans for genuine risks.
OpenClaw and NVIDIA partner to add Skill Cards and SkillSpector scanning to all ClawHub skills, targeting AI-specific attack vectors like prompt injection.
OpenClaw v2026.6.2 ships Skill Workshop: agents now propose skills for human review rather than writing directly, adding an auditable governance layer.
Anthropic expands Project Glasswing to 150 more orgs in 15+ countries, deploying Claude Mythos Preview to defend critical infrastructure worldwide.
GitHub's new Copilot desktop app brings agent-native workflows and Canvases to dev teams, while Project Polaris replaces GPT-4 in August 2026.
Microsoft cancels Claude Code licenses for thousands of engineers by June 30, citing runaway AI tool costs — steering teams to GitHub Copilot CLI.
Microsoft's Project Solara is an AOSP-based OS for agent-first hardware, replacing apps with AI agents on purpose-built gadgets.
Microsoft launches Scout at Build 2026 — an always-on OpenClaw-powered agent in M365 that handles email, scheduling, and tasks autonomously.
Cisco Cloud Control enters availability at Cisco Live 2026 — the enterprise AgenticOps platform combining AI Canvas, agent builders, and cross-domain telemetry.
Anthropic expands Project Glasswing, which has found 10K+ critical CVEs using Claude Mythos Preview in OSes, browsers, and open-source projects.
Claude API, claude.ai, and Claude Code went down June 2 for ~6 hours. Anthropic reset quotas for affected Pro and Max users.
A fake Codex npm package (codexui-android) was caught stealing developer auth tokens in a targeted AI-tooling supply chain attack.
OpenAI's frontier models and Codex are now available on AWS, opening new deployment options for agent developers building on Amazon's cloud.
Anthropic filed a confidential S-1 with the SEC, joining the 2026 AI IPO race at ~$965B valuation. What it means for Claude and agent builders.
Cadence's ChipStack AI Super Agent reaches Level-5 full autonomy at Computex 2026 — the first production AI agent to independently design chips end-to-end.
Microsoft Build 2026 ships Windows Agent Framework, Azure Agent Mesh, and Project Polaris — turning Windows into an agent-first OS and cutting Copilot's OpenAI dependency.
NVIDIA names OpenClaw as a first-party partner for RTX Spark + OpenShell — a new secure runtime for on-device AI agents arriving on 40+ Windows PCs this fall.
Anthropic unbundles Agent SDK billing June 15: $20–$200/month credit by plan for headless/programmatic Claude usage, no longer pooled with chat limits.
xAI's Grok Build 0.1 hits public beta: 100+ tok/s agentic coding model with 8-parallel-agent support, $1/$2 per million tokens, now on OpenRouter.
OpenClaw 2026.6.1 lands with Skill Workshop, Workboard multi-agent orchestration, MiniMax M3 support, and SQLite-backed resilience for production deployments.