Google has signed an agreement with the US government allowing the Pentagon to use its AI models for classified work. The contract permits Google's AI to be used for what it describes as any lawful government purpose, while explicitly stating the system should not be used for domestic mass surveillance or autonomous weapons without appropriate human oversight. It also notes that the agreement does not give Google any right to control or veto government operational decisions made using the technology.

The deal was finalized despite internal opposition from Google employees, which is a notable detail given the company's history. In 2018, Google walked away from Project Maven, a Pentagon drone program, after employees revolted and thousands signed an open letter demanding the company stay out of warfare AI. That revolt was widely seen as a defining moment for the idea that tech workers could shape their employer's ethics. Today's news suggests the window for that kind of influence has narrowed considerably.

The context has changed too. The defense sector's appetite for AI has grown dramatically since 2018, and every major competitor, including Microsoft and Amazon, has been aggressively pursuing government contracts. Staying out of that market entirely becomes harder to justify to shareholders when adversaries and rivals are moving in.

Whether the safeguards written into the contract hold up in practice is the question nobody can answer today. The agreement says the right things about human oversight. It also says Google has no veto over how the government actually uses it. Those two things will eventually be in tension, and when they are, it will not be Google making the call.

Two updates dropped today that point in different directions but land the same message: Claude is becoming the connective tissue between AI and the tools people already use every day.

The first is practical and immediately useful for developers. Claude Code can now send push notifications to your phone when a long task finishes or when it needs your input. You set it up by pairing your phone via QR code, then walk away from your computer entirely. Claude handles the work and texts you when it is done or stuck. For developers running long test suites or complex builds, this changes the workflow entirely. You are no longer waiting at your desk watching a progress bar.

The second update is bigger in scope. Anthropic launched a set of connectors that plug Claude directly into the creative tools professionals actually use: Blender, Autodesk Fusion, Adobe Creative Cloud, Ableton, Splice, Canva, SketchUp, and Resolume. In plain terms, this means a 3D designer can now describe changes to a model in conversation and have Claude execute them inside Autodesk. A music producer can work with Claude inside Ableton. A motion designer can batch-apply changes across an entire scene in Blender through natural language. Anthropic also joined the Blender Development Fund as a patron, signaling this is not just a product integration but a longer-term bet on the creative software ecosystem.

Taken together, these two updates are part of the same strategy. Claude is not trying to replace the tools creative professionals and developers rely on. It is trying to become the layer that makes all of them faster.

A single entry labeled GPT-5.6 was spotted in Codex backend logs, buried among routing data that otherwise pointed to GPT-5.5. One single log entry most likely indicates a canary test, where a small slice of traffic gets quietly routed to an experimental build. It is not evidence that GPT-5.6 is shipping soon.

What it does tell you is that the pace has not slowed down. GPT-5.5 is barely out the door and something is already being tested behind it. Worth keeping an eye on, not worth reading too much into yet.