June 16 is dominated by the widening gap between how fast AI capability is shipping and how fast the guardrails around it are catching up. Anthropic was ordered by the US government to cut off foreign access to its flagship Fable 5 and Mythos 5 models on national-security export-control grounds, while security researchers at Check Point disclosed a SQL-injection-to-RCE chain in LangGraph's checkpointer, showing agent frameworks are now inheriting classic web-app bugs with much higher blast radius. On the infra side, NVIDIA's Blackwell topped the first dedicated agentic-AI benchmark (AgentPerf), Moonshot shipped an open-weights Kimi K2.7 Code model, and GitLab and Ramp both pushed further into agent-native tooling and finance-specific coding benchmarks. Rounding things out, Adyen is buying billing startup Orb for $335M, Oracle quietly shrank its Always Free cloud tier, and Apple continues its low-key AI rollout with new design tooling and under-the-hood Siri plumbing it chose not to demo at WWDC26. For engineers, the throughline is clear: agent frameworks and frontier model access are both becoming regulated, exploitable surface area, not just productivity tools.

• Anthropic disabled Fable 5 and Mythos 5 for foreign users under a US export-control order.
• Check Point found a SQLi-to-RCE chain in LangGraph's checkpointer component.
• NVIDIA Blackwell topped the first agentic-AI infrastructure benchmark, AgentPerf.
• Moonshot AI released Kimi K2.7 Code as open weights for agentic coding.
• Ramp launched a private SWE-Bench variant just for financial coding agents.