Subject: AI Signals: New Models, New Privacy Risks

This week’s AI Signals post tracks a sharp shift in the stack: Google launched Gemini 3.1 Flash Live and expanded Lyria 3, AWS opened Bedrock from New Zealand, GitHub changed Copilot interaction-data defaults ahead of April 24, and Mythos/Capybara leak reporting reinforced how frontier models are now being staged behind tighter safety controls. The pattern is clear: capability is accelerating, but policy, power, and release governance are now the real bottlenecks.

That is exactly why today’s second post matters: “The Digital Butler or Trojan Horse?” translates those headlines into a practical privacy playbook for persistent agents, with concrete code for policy-gated tool execution, untrusted-input sanitization, and isolated runtime boundaries. If you are building or using AI agents right now, this is the piece that helps you move fast without handing over control.