GPT-5.6 Government Freeze: From Model Wars to Context Wars

Key Takeaways

• GPT-5.6 isn't cancelled — it's gated. Access is restricted to a small group of government-approved partners while Washington reviews cybersecurity risk. That slows public frontier availability without stopping the labs' private, internal progress.
• The "context war" is the through-line. Jones's core claim: the next useful AI product won't be the one that wins a benchmark — it'll be the one that knows where your work lives, what it's allowed to see, and what it's allowed to do, seamlessly.
• Claude Tag brings a teammate into Slack. Anthropic's launch lets a team grant Claude scoped access to selected channels, tools, data and code bases, with permissions, spend limits and channel-defined memory — a bet on messy, informal team context after formal context via Claude Code and Cowork.
• Apple's Siri is a "context answer." The Siri relaunch connects on-device to calendar, photos, email, notes and app state (on-device first, private cloud when needed) so a modestly-smart model becomes very useful — capability may be the wrong question.
• Codex had to earn trust — even inside OpenAI. The adoption study shows non-engineers only fed Codex legal/recruiting/sales context after it proved itself; adoption jumped after 5.5. Jones frames Codex as "file-shaped" (point it at your files) versus Claude as "chat-shaped" (tag it into the conversation).
• Frontier friction helps open source. Slowing public frontier releases gives open models (GLM 5.2, and whatever DeepSeek/Qwen ships next) room to close the public gap even if labs keep a 6–8 month private lead — raising the value of owning your own context and harness.