OWASP: Prompt Injection Is a Permanent Architectural Flaw, Not a Patchable Bug
OWASP's 2026 agentic AI report says prompt injection cannot be patched — it's baked into how LLMs process tokens, requiring architectural design changes.
OWASP's 2026 agentic AI report says prompt injection cannot be patched — it's baked into how LLMs process tokens, requiring architectural design changes.
AWS AgentCore gets a dedicated AG-UI endpoint and FAST template, making one-command production agentic web app scaffolding with CopilotKit a reality.
Microsoft's Azure Container Apps Sandboxes (public preview) offer Hyper-V microVM isolation for untrusted AI agent code — fast, scalable, and enterprise-grade.
Huawei's HarmonyOS 7 ships with 2,000+ AI agents and a system-level XiaoYi assistant claiming >90% task success — a major non-US agentic deployment.
Trump AI adviser David Sacks blames Anthropic's refusal to patch a jailbreak for triggering export controls on Fable 5 and Mythos 5 — calling it 'recklessness.'
A detailed hour-by-hour account of how the White House escalated from concern to export-control directive on Anthropic's Fable 5 and Mythos 5 in 24 hours.
CrewAI v1.14.7 stable lands Chat API for Flows, route-aware decorators, and Snowflake Cortex provider support — production-ready after June alphas.
Gartner's 2026 Hype Cycle shows only 17% of orgs run live AI agents despite 60%+ planning near-term deployment — and 40% of projects may fail by 2027.
MCP's upcoming stateless core redesign removes session IDs and handshakes, enabling horizontal scale — RC due July 28 with breaking changes ahead.
OpenClaw 2026.6.8-beta.1 ships richer Telegram/WhatsApp delivery: structured rich text, ACP bindings, safer media boundaries, and security fixes.
Bastion launches on HN as an isolated Linux VM layer for running background coding agents safely, addressing supply-chain and lateral-movement risks.
Claude Code v2.1.175 adds enforceAvailableModels: enterprise teams can now lock the model allowlist at the org level, blocking user overrides for compliance.
Databricks open-sources Omnigent, a meta-harness letting teams compose and control multi-agent systems across Claude Code, Codex, and custom agents.
OpenClaw 2026.6.8-beta.1 adds Telegram structured rich text (tables, lists, blockquotes) and WhatsApp ACP binding support for richer agent delivery.
Tenet Security's 'agentjacking' attack hijacks AI coding agents via fake Sentry error reports — no malware or phishing needed. 100+ orgs tested.
MCP 2026-07-28 RC lands stateless protocol core, Extensions framework, and OAuth 2.1 hardening — the biggest spec change since launch. Final: July 28.
OpenClaw v2026.6.7-beta.1 ships Kimi K2.7 Code support, Feishu context-leak fix, WebSocket hardening, and channel delivery improvements.
StakeBench study: no AI web agent reliably blocks prompt injection. GPT-5 and Gemini both fail — direct attacks succeed over 79% of the time.
Anthropic suspended Claude Fable 5 and Mythos 5 globally after a U.S. government export-control directive — just three days after launch.
Google DeepMind and Schmidt Sciences launch a $10M research call targeting emergent safety risks when millions of AI agents interact at scale. Deadline Aug 9.
Moonshot AI open-sources Kimi K2.7-Code — a 1T-parameter MoE coding agent with 256k context, tool calling, and commercial-friendly Modified MIT license.
Visa and Mastercard both launch dedicated AI agent payment rails in June 2026 — marking the moment autonomous agents can complete real financial transactions.
Anthropic splits programmatic Claude Code usage to separate monthly credits on June 15. Action required before deadline for CI/CD users.
Claude Code v2.1.172 unlocks 5-level nested sub-agent hierarchies, fixes 1M-context sessions, and adds Bedrock region auto-detection.
Hackathon winner open-sources 50+ skills and 13 specialist agents in one Claude Code production starter kit.