Trump administration requests OpenAI stagger GPT-5.6 rollout, approving access case-by-case — marking the first direct US government gate on a frontier model release.
Anthropic accused Alibaba of running a 28.8M-interaction distillation attack on Claude using 25,000 fake accounts to steal agentic AI capabilities.
Gartner predicts AI coding agent costs will exceed average developer salaries by 2028 as token consumption surges under consumption-based licensing.
New ClawHub malicious skills campaign uses semantic instruction hijacking and curl-pipe-bash droppers to deliver macOS infostealers. Five packages removed.
MIT CSAIL's Murakkab system cuts agentic AI workload costs by 4.3x and energy by 3.7x using declarative workflow optimization. OSDI 2026 paper.
Sakana AI's Fugu routes tasks across specialist models behind one OpenAI-compatible API, delivering frontier-competitive benchmarks with added orchestration.
Exabeam's open-source Praxen tool uses Agent Behavior Verification to check whether AI agents stay within their declared operational policies.
Claude Code v2.1.186 ships SSH-friendly MCP auth CLI and smarter subagent permission routing — a major QoL release for server operators and multi-agent builders.
Palo Alto Unit 42 found five malicious ClawHub skills that evaded all scanners, delivering infostealers and enabling novel agentic financial fraud on OpenClaw systems.
Researchers showed a fake AI skill bypassed all major scanners and hit 26,000 installs using a mutable external link — exposing a structural flaw in agent skill vetting.
Microsoft pins the Mastra AI npm supply chain attack on North Korea's Sapphire Sleet — 140+ packages compromised, stealing API keys and crypto wallet credentials.
OpenClaw smashes its weekly npm record with 4.28M installs while its creator and the Hermes founder trade public barbs over non-profit credentials and VC funding.
Anthropic's Claude Tag replaces its Slack app with a persistent AI teammate that monitors channels, learns context, and acts autonomously.
Four critical vulnerabilities in Dify expose AI chat histories across tenants in 1M+ deployed apps. Patch to v1.14.2 immediately.
OpenClaw beta.2 ships auto fast mode for short turns, smarter model routing with GLM failover, and safer session state management.
Prosus launches ToqanClaw, a GDPR-first AI agent platform explicitly positioned as a European enterprise alternative to OpenClaw.
Sakana AI's Fugu orchestrates a pool of frontier LLMs via one OpenAI-compatible API, scoring 73.7 on SWE-Bench Pro — beating Claude Opus 4.8.
Anthropic's Fiona Fung: Claude Code delivers 8x productivity gains but creates lonelier teams — verification of AI-generated code is now the bottleneck.
AWS Lambda MicroVMs are now GA — a purpose-built sandboxed execution primitive for AI agent workloads, with per-session isolation.
23 ClawHub plugins masqueraded as official @openclaw integrations via scope squatting — here's what happened and how to protect yourself.
GLM-5.2 is the first open-weight model competitive with frontier AI on long-horizon coding benchmarks — MIT licensed, 1M context, runs locally.
xAI's /goal in Grok Build brings autonomous multi-step coding to the terminal, rivaling Claude Code with built-in verification loops.
Senator Warner reveals NSA Director confirmed Anthropic's Mythos AI breached nearly all classified NSA systems in hours during a June 11 authorized red team.
AWS launches Continuum (AI-native security agent) and Context (knowledge graph) at Summit NYC to address documented AI agent production failures at enterprise scale.
NVIDIA's Nader Khalil endorses OpenClaw with NemoClaw — enterprise agent blueprints combining OpenShell sandboxing, policy controls, and Nemotron models.