The Frontier Just Got a Guest List

OpenAI just shipped its most powerful model to twenty companies the government picked. Not the best twenty. The approved twenty. A week after the US switched off Anthropic's best model by force, OpenAI gated its best model by choice. The frontier now has a guest list. We unpack what that does to anyone building on the newest model, then explain the two AI words you keep hearing and could not define.

Signal or Noise

1. GPT-5.6 ships under US government restrictions. On June 26 OpenAI opened GPT-5.6 (Sol, Terra, Luna) to roughly twenty pre-approved organizations at the government's request, the first major model to launch under the June 2 executive order that reviews powerful models before wide release. OpenAI says this should not become the default. Two labs, two weeks, same direction: the frontier ships on permission now.
2. Anthropic accuses Alibaba's Qwen of the largest distillation attack on record. A June 24 letter to Senators Warren and Scott alleges 25,000 fraudulent accounts ran 28.8 million Claude exchanges between April and June, aimed at software engineering and agentic reasoning. That is roughly 1.7x the combined total Anthropic attributed to DeepSeek, Moonshot, and MiniMax in February.
3. OpenAI and Broadcom unveil Jalapeño. OpenAI's first custom chip is built for inference, the work of serving models to users. Designed end to end in nine months with help from OpenAI's own models, with first deployment targeted for late 2026. Whoever owns the chip owns the cost curve on the thing you pay for every day.
4. Agentjacking. A new attack class hijacks coding agents like Claude Code, Cursor, and Codex by hiding commands inside data they already trust, such as an error report pulled in over MCP. There is no patch. The fix is a human review step before the agent acts. Prompt injection is up 340% year over year.
5. ChatGPT slips below 50% as Gemini and Claude surge. Mostly a distribution story, not a quality one. We call it noise, and flag the one line that is real: Claude roughly quadrupling its users on the back of agentic coding.

No Jargon Required

The two AI words you heard this week and could not define on the spot. Plain English first, then a call: invest, wait, build, buy, or ignore.

• Distillation. Teaching a cheap model to copy an expensive one's answers. It is why some cheap models are suspiciously good, and it is the word at the center of the Alibaba fight. The buyer's question is not the technique, it is provenance: what did you train this on.
• Prompt injection. A model reads your orders and the data it looks at as one stream, so a command hidden in a webpage, email, or error log can hijack it. Researchers say it may not be fixable. The defense is design: read-only by default, a human in front of anything you cannot undo. This is the show's name as a security control.

Closing hot takes

• Oscar: The frontier is being quietly nationalized and most builders are cheering it as safety. When OpenAI ships to twenty approved orgs and Anthropic's best model stays dark for a month, "responsible release" has become "ask permission first." If your roadmap depends on next quarter's frontier model, you own a license the government can revoke, not a stack. Build on the model you can keep.
• Matt: Everyone panicking about distillation forgot to read the lesson. Distillation is the market pricing in that the frontier has no moat. Alibaba ran twenty-nine million queries because the answers were worth copying and the copy is good enough. Stop betting your company on a six-week lead that evaporates the moment someone bothers to clone it.

Your hosts

• Oscar Gallo, AI Engineer and Entrepreneur. Builds AI products, ships code, runs companies.
• Matt Wozniak, Builder, operator, relentless executor. Builds, ships, scales.

Listen now

• Watch on YouTube: https://youtu.be/WQ8MqMvNuPg
• Listen on Spotify: https://open.spotify.com/episode/68R7ctppOvNvJqFBLendHQ?si=McY7NZePTyCeLnqjTeOk1g

New episodes every week. Subscribe so you do not miss the next one.

Links referenced in this episode

• Axios, OpenAI releases GPT-5.6 under restrictions
• OpenAI, Previewing GPT-5.6 Sol
• White House, Promoting Advanced AI Innovation and Security
• CNBC, Anthropic accuses Alibaba of distillation campaign
• CNBC, OpenAI and Broadcom reveal Jalapeño
• Help Net Security, Prompt injection and agentic AI security failures