OpenAI lists its GPT-6-Cyber model at $20 per million input tokens and $100 per million output tokens, twice GPT-6 Astra’s $10 and $50 by our arithmetic, and requires Daybreak Red access and approval to use it (our piece).
In the same week, the Frontier Model Forum released version 1.0 of a Jailbreak Rating Framework for Frontier Cyber Threats, with executives from five companies quoted backing it: Amazon Web Services, Anthropic, Google DeepMind, Microsoft and OpenAI (our piece). By our count, all five sell access to frontier AI models through their own products.
Business Pill · AI Agents Explained
A short explainer on AI agents, systems that move from answering to doing. It teaches the general idea only and does not discuss this story.
The key insight: In one week OpenAI priced GPT-6-Cyber at twice GPT-6 Astra by our arithmetic, five companies that sell frontier-model access backed a scale for rating cyber jailbreaks, and model makers ran four of eight published agent studies, by our count.
First, the price. OpenAI’s GPT-6-Cyber page says “Your organizationneeds Daybreak Red access and approval to use GPT-6-Cyber.” By our arithmetic, a job using one million input and one million output tokens costs $120 on GPT-6-Cyber and $60 on GPT-6 Astra at listed short-context rates (our piece).
Second, the scale. The framework scores a successful jailbreak on four categories, each from 0 to 3, and turns the result into one of five risk bands, from Negligible to Critical. The forum calls it preliminary and has opened public comment until 21 December 2026 (our piece).
Third, the studies. Of eight AI agent studies published on 5 and 6 November, four were run by the company that makes the model being tested, on its own models, by our count: two by Anthropic, one by OpenAI and one by Google DeepMind. The other four came from outside groups: Visa, Snorkel AI, METR and Physical Superintelligence (our piece).
Who backed this week’s cyber jailbreak scale and who ran the eight AI agent studies. Source: our count from FourWeekMBA coverage, 5-7 November 2026. Classification: ours.
Our Numbers
By our count: 5 of the 5 companies quoted backing the jailbreak scale sell access to frontier models, and 4 of the week’s 8 agent studies were run by the model maker on its own models. The cyber model’s list price is 2x the flagship’s on both input and output, by our arithmetic.
The Structural Read
As we read these three pieces together, the companies that sell frontier models are, this week, also the main authors of the scale for rating cyber misuse of those models and of half the published agent evaluations. That is a description of who wrote what, not a finding about the quality of any of it. The framework is open for public comment, and four of the eight studies came from outside groups.
The Structural Read
As we read our own coverage, the same group of companies appears on three sides this week: as the seller of a cyber model, as backers of the scale that rates cyber jailbreaks, and as the runners of half the agent evaluations.
As we read it, the outside checks are also visible: the framework is open for public comment until 21 December 2026, and four of the eight studies came from Visa, Snorkel AI, METR and Physical Superintelligence.
As we read it, this is a map of who wrote what this week, not a judgement on the quality of the pricing, the scale or the studies.
“Your organizationneeds Daybreak Red access and approval to use GPT-6-Cyber.”
Three Implications
FOR BUYERS OF CYBER AI The listed price is 2x the flagship’s and access needs approval, by OpenAI’s terms; the scale that may be used to rate jailbreaks of such models is still preliminary.
FOR READERS OF AGENT BENCHMARKS Half of this week’s studies were run by the maker on its own models, by our count, so who ran each study is worth checking before comparing results.
FOR THE PUBLIC COMMENT WINDOW The Frontier Model Forum framework takes comments until 21 December 2026, per our earlier piece.
The analysis says: “The surface, not the model, governs autonomy.”
As we read it, this week’s cyber pricing, jailbreak scale and agent studies are all set at the surface around the models, where access, rating and testing happen; the analysis does not discuss these stories.
What Is Not Established
Our pieces do not establish whether GPT-6-Cyber has been rated on the forum’s scale, how many organisations hold Daybreak Red approval, or whether any study’s result would differ if run by an outside group. The studies measure different things, so their scores are not comparable with each other.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
In one week OpenAI priced its cyber model at twice its flagship, five model-selling companies backed a scale for rating cyber jailbreaks, and model makers ran four of eight agent studies, by our count.
A note on sourcing. This analysis draws only on our own earlier coverage, linked in the text; the counts and classifications are ours, and the price ratio is our arithmetic from OpenAI’s listed prices.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
