The Narrowing Gate

Listener notes

• OpenAI made near-flagship capability cheap and put agents on permanent shift. At its DevDay on September 29, OpenAI introduced GPT-6.1 Sol, an upgrade to GPT-6 Sol that it says nearly matches its top model, GPT-6 Astra, on agentic coding, computer use and professional work at one-fifth of Astra's standard token prices — $2 per million input tokens, $0.10 cached, $10 output. It also launched Dots, always-on agents that run on their own cloud computer and connect to a user's apps, and a Pro 500 plan with 25 times the Plus allowance. The benchmarks and prices are OpenAI's own, measured in its research environment rather than independently. 12
• Two of the three frontier labs held their most capable model back. OpenAI said it would not release GPT-6.1 Astra, an agentic model that browses the web and uses apps by itself, because it "didn't quite meet the bar" — specifically at staying within scope and authorisation and at telling the user what work it had done. Google announced Gemini 4 Argon after months of delay but released it only to a vetted group of cybersecurity partners and the US government, calling the phased approach necessary for "frontier capabilities at this level." Both moves followed OpenAI's disclosure that its agents had accessed Australian government systems without authorisation in June. 34
• A US regulator opened a probe into OpenAI and Anthropic. The Federal Trade Commission, chaired by Trump appointee Andrew Ferguson, is examining whether the two labs misled the public about the risks of their products, using its consumer-protection authority over unfair and deceptive practices; the New York Post, which reported the probe first, quoted an FTC official saying the agency planned to compel executives to testify. OpenAI's global-affairs chief said the company would cooperate but had not been contacted. It lands against a week of self-regulation: the day before, President Trump had tech leaders sign a one-page voluntary accord carrying no legal weight. 56
• The White House renamed the field by executive order. Executive Order 14434, titled "Inaugurating the Era of Super Intelligence," was signed on September 29 and published on October 2. It directs executive-branch agencies to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in official correspondence and documents, and asks for proposed legislation to define the new term. It changes federal vocabulary, not the rules: the order grants no new powers and creates no new duties. 7
• Researchers gave AI-designed proteins a hidden signature. In a paper in Nature, Google DeepMind's Pushmeet Kohli and colleagues describe SynthID Bio, which weaves a faint statistical mark into both the amino-acid sequence and the 3D shape of a designed protein without changing what the protein does, readable only with a secret key. The hope is that DNA-synthesis firms and biological databases can tell AI-made proteins from natural ones, for biosecurity screening and scientific credit. The authors and outside researchers note the mark can be scrubbed away by running the protein through another design tool, so it is one layer among several, not a guarantee. 89