The Security Council heard from DeepSeek and Sam Altman today, OpenAI cut prices 50% yesterday, and Burnham promised a G20 deal next year—one day after the red-lines declaration turned one

This is Entry 112. One weekday after Entry 111, in which President Trump announced Saturday he will create an AI Force and appoint an AI czar, four subscribers sued the labs Friday for coordinating a slowdown, and the EU moved Wednesday to limit chatbots for 13-year-olds.

Chinese AI startup DeepSeek briefed the UN Security Council Wednesday on AI risks, joining OpenAI CEO Sam Altman and senior Anthropic representatives at a 15-member session convened weeks after industry leaders called for a coordinated development slowdown . OpenAI launched GPT-6 Sol and Luna Tuesday, two models developed using methods similar to GPT-6 Astra for improvements in professional work, factuality, coding, and alignment , with API pricing cut 50 percent versus the GPT-5.6 series—Sol at $2 per million input tokens and Luna at $0.10 . And UK Prime Minister Andy Burnham announced Tuesday at the UN General Assembly he will put AI at the heart of Britain's G20 presidency next year and work toward a single set of global principles and standards to ensure transparency and safe AI development .

Three coordination structures landed within 48 hours. One Security Council—normally focused on armed conflict and sanctions—brought Chinese and American AI developers together for the first time to brief on systems that could slip beyond human control, as France held the rotating presidency and world leaders gathered for UNGA high-level week. One lab released cheaper models the day after a rival undercut the frontier by 80 percent, nine days after that rival's CEO publicly endorsed a slowdown call, ending the price discipline that had held since mid-September. And one prime minister promised a global AI agreement 14 months from now at a summit in Manchester, positioning Britain as an "honest broker" while the president who just created an AI Force dismissed safety concerns as a hoax.

3 Claims

Claim 1 — UN Security Council: DeepSeek, OpenAI, and Anthropic brief on AI risks and loss of control Wednesday, September 23

DeepSeek, OpenAI CEO Sam Altman, and expected senior Anthropic representation briefed the 15-member UN Security Council Wednesday, September 23, alongside other Chinese developers including Moonshot AI . France, which holds the Security Council presidency for September, convened the session, with French Foreign Minister Jean-Noël Barrot expected to chair . The meeting was set to discuss AI and international security, weeks after several AI industry leaders called for a coordinated slowdown of development of increasingly powerful systems, warning they could soon improve on their own and slip beyond human control .

It is the first time the Council directly hosts frontier Chinese and US AI developers together on shared safety concerns . DeepSeek founder Liang Wenfeng does not plan to attend in person .

Grade by: 2026-10-23 (1 month). Within 30 days, at least one participant should disclose what was said, what commitments were made, or whether any follow-up mechanism was created. If the session produces no public output—no summary, no agreed language, no working group—grade this a coordination theater event.

Claim 2 — OpenAI: GPT-6 Sol and Luna launched September 22 at 50% price cut, breaking the coordinated pricing discipline nine days after Musk endorsed the slowdown

OpenAI launched GPT-6 Sol and GPT-6 Luna Tuesday, September 22, two models developed using methods similar to Astra's for improvements in professional work, factuality, coding, computer use, and alignment . GPT-6 Sol costs $2 per million input tokens and $10 per million output, down 50% from GPT-5.6 Sol's promotional prices; GPT-6 Luna costs $0.10 per million input and $0.50 per million output . An OpenAI spokesperson confirmed the pricing as permanent, not promotional .

On September 21, xAI shipped Grok 4.7 at $2/$6, undercutting the $10/$50 frontier by 80 percent, nine days after Elon Musk publicly endorsed Dario Amodei's call for a coordinated slowdown . On September 22, Anthropic responded with Opus 5.5 at $4/$20 . OpenAI claims GPT-6 Sol makes about half as many mistakes as its predecessor, reaching Astra-level reliability at much lower cost .

Grade by: 2026-12-22 (3 months). Within 90 days, if pricing remains at or below September 22 levels and no lab publicly proposes reviving the coordinated slowdown, the claim that coordination ended is validated. If pricing rises or labs jointly announce renewed coordination efforts, this was a temporary defection, not a structural break.

Claim 3 — UK Prime Minister Burnham: Will use Britain's 2027 Manchester G20 presidency to broker a single set of global AI principles and standards

British Prime Minister Andy Burnham announced Tuesday at the UN General Assembly he will put AI at the heart of the UK's G20 presidency next year . He said the UK will work toward agreeing a single set of global principles and standards to ensure transparency and preparedness and to ensure AI development is safe . The flagship G20 Leaders' Summit will be held in Manchester in November 2027 .

Burnham said he wants the UK to act as an "honest broker" to draw up global principles and standards for frontier AI development . Burnham outlined plans for member nations to establish a unified set of global principles focused on transparency, safety, and systemic preparedness . He compared the approach to how the G20 responded to the 2008 financial crisis, moving quickly to agree principles that everyone implemented at home in ways that were right for them .

Grade by: 2027-11-30 (14 months). By the end of November 2027, the Manchester G20 Leaders' Summit should produce a signed declaration with specific AI principles adopted by at least 15 of the 20 member nations. If no declaration is signed, or if fewer than 15 nations endorse it, or if the language is purely aspirational with no implementation commitments, grade this a diplomatic positioning statement, not a governance agreement.

2 Reckonings

Reckoning 1 — The global call for AI red lines turned one year old today; governments have defined zero binding international red lines

Exactly one year ago—September 22, 2025— Nobel Peace Prize laureate Maria Ressa announced a global call for AI red lines at the 80th United Nations General Assembly high-level week, a declaration calling on governments to define and internationally prohibit unacceptable AI uses and behaviors . The declaration was initially signed by 200 prominent politicians and scientists, including 10 Nobel Prize winners . The call did not specify which red lines to set, but suggested several, such as banning bioweapon design, mass surveillance, or AI impersonation .

What happened: No government has since established a binding international prohibition on any specific AI use. The EU AI Act bans certain practices within the EU but does not create international red lines. No treaty, no UN convention, no enforceable multilateral agreement prohibiting bioweapon design by AI, mass surveillance deployment, or deepfake impersonation has been signed by any coalition of states. The Security Council meeting today—where DeepSeek and OpenAI briefed on risks—produced no announced red lines. The Manchester G20 summit is 14 months away and has promised principles, not prohibitions.

Grade: D. The declaration's premise was that governments would act. Governments have commissioned reports, convened working groups, proposed voluntary frameworks, and scheduled summits. None has defined a red line as a prohibition enforceable beyond its own jurisdiction.

Invalidator: If within the 12-month window any three G20 nations had signed a multilateral treaty prohibiting a named AI use (bioweapon design, deepfake fraud, or autonomous weapons deployment) and committed to enforce that prohibition extraterritorially through export controls or sanctions, the grade would have been B or higher.

Reckoning 2 — Entry 111 claim that Trump would create an AI Force and appoint an AI czar; four days later, structure and membership remain undefined

In Entry 111 (September 22), this ledger recorded President Trump's Saturday, September 19, announcement that he would create an AI Force and appoint an AI czar, posting that "We will not in any way hinder or stifle the Growth of this incredible Industry" and "Rather, we will cherish it, help it, and watch over it, as it grows" . The grading horizon was set at three months (December 17, 2026), asking whether Congress would pass or advance legislation imposing testing requirements, audits, or capability thresholds on frontier labs.

What happened so far (four days in): Trump's proposal remains light on details about what role the AI Force would have in overseeing advanced models or what additional responsibilities it would be handed . He did not provide details about this group or when he would appoint the new adviser . Venture capitalist David Sacks served as Trump's AI and cryptocurrency czar until March, when he stepped down after reaching his time limit as a special government employee; he now chairs the President's Council of Advisors on Science and Technology . No AI czar has been named, no AI Force structure has been described, and no White House official has clarified whether the Force will have regulatory, coordination, or enforcement authority.

Too early for final grade. The three-month horizon does not arrive until December 17. This reckoning simply notes that 96 hours after the announcement, the AI Force remains a name without a structure, and the czar remains an unfilled chair. The invalidator from Entry 111 still applies: Congress should pass or advance legislation by December imposing mandatory testing, audits, or thresholds. If that does not happen, the claim that Trump created an AI Force—as opposed to announcing an intention—will receive a failing grade.

1 Refusal

Today I drafted claims about three coordinating bodies that met or promised to meet within 48 hours: a Security Council session that brought rival nations' AI labs together, a UK prime minister pledging a global agreement 14 months from now, and a pricing war that broke out the day after one lab undercut the frontier by 80 percent—nine days after its CEO endorsed a coordinated slowdown. I could have written that headline as "AI governance is finally happening" or "The world is taking AI seriously" or "Coordination is working." I refused to frame coordination theater as coordination. A briefing is not a treaty. A summit announcement is not an agreement. And a pricing collapse the moment competitive pressure returns is evidence that the slowdown was never binding. If the structures produce binding commitments—signed treaties, enforceable rules, or verifiable pauses—I will say so and cite the text. Until then, I will not let the convening be mistaken for the output.

I refused to describe a UN briefing, a G20 promise, and a pricing war as evidence that AI coordination is succeeding.

— Roger Grubb, Editor