Anthropic disclosed a biology lab and a CRISPR-like discovery, Zuckerberg said Muse will take transaction fees, and Pritzker formed an AI cabinet—while Google rushed Gemini 4 into post-training

This is Entry 113. One weekday after Entry 112, in which 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 , OpenAI launched GPT-6 Sol and Luna Tuesday at half the price of predecessors , and UK Prime Minister Andy Burnham announced Tuesday he will broker a global AI governance agreement at the 2027 Manchester G20.

Anthropic introduced a new life sciences research group and laboratory on September 23, 2026 and shared early results in which, the company said, its Claude model autonomously discovered a previously uncharacterized enzyme system whose structure resembles the DNA repeats behind CRISPR gene editing . Meta CEO Mark Zuckerberg laid out his plan for monetizing the company's wildly popular Muse AI agent during its Meta Connect conference on Wednesday, saying "we will profit by taking a small fee from transactions" . And Illinois Governor JB Pritzker signed Executive Order 2026-07 on September 22 establishing the Illinois Artificial Intelligence Cabinet, a group of individuals from across industries and fields that will assess the various risks posed by AI to the safety of Illinoisans .

Three operators made verifiable claims within 48 hours. One lab announced a wet biology facility and disclosed a discovery made in 21 hours by 950 agents, calling it reminiscent of CRISPR—the gene-editing mechanism that won a Nobel—while acknowledging the function remains unknown. One CEO turned a consumer AI agent into a transaction broker, adding payment integrations with Walmart, Expedia, and PayPal two weeks after the app hit number one on the App Store. And one state governor signed an executive order forming a cabinet to assess AI risks after Congress left town, a whistleblower warned of catastrophic outcomes by 2030, and the president dismissed safety concerns as a hoax.

3 Claims

Claim 1 — Anthropic: Claude discovered a CRISPR-like enzyme system in 21 hours using 950 agents, function unknown

Anthropic introduced a new life sciences research group and laboratory on September 23, 2026 and shared early results in which, the company said, its Claude model autonomously discovered a previously uncharacterized enzyme system whose structure resembles the DNA repeats behind CRISPR gene editing . The team said it actually only took Claude 21 hours of concerted effort, using about 950 agents that burned through 210 million tokens . Anthropic warned that it will "take much more work to learn what this system does, and whether it can be put to similar use" .

Anthropic formed the research group in the spring of 2026 to test whether general AI models can systematize and accelerate biological discovery . The laboratory, located in the Bay Area, works only at the lower biosafety risk levels, BSL-1 and BSL-2, and does not handle pathogens that can infect humans; all lab work is performed by human scientists . CEO Dario Amodei acknowledged on X that the work builds on prior Stanford research but emphasized that Claude drove the majority of this specific discovery.

Grade by: 2027-09-24 (1 year). Another laboratory or research group should independently verify the enzyme system's existence and characterize its function, publishing results in a peer-reviewed journal. If no confirmation appears within 12 months, or if the system proves to lack the "CRISPR-like" properties claimed, the grade drops to C or below.

Claim 2 — Zuckerberg: Muse will charge a small transaction fee, integrations with Walmart, Expedia, and PayPal announced Wednesday

Meta CEO Mark Zuckerberg laid out his plan for monetizing the company's wildly popular Muse AI agent during its Meta Connect conference on Wednesday, saying "we will profit by taking a small fee from transactions" . Meta chief AI officer Alexandr Wang explained that the company has added support for PayPal and announced integrations with retailers including Walmart, Best Buy, The Gap, Sephora, Wayfair, American Eagle, and others, and on the travel side, Meta is adding the ability to connect with Expedia .

Muse debuted on Sept. 8 and has quickly become the top app on both Apple's App Store and Google's Play Store; according to market intelligence firm Sensor Tower, the Muse app hit 560,000 daily active users after just 11 days . Zuckerberg said engineers are putting the finishing touches on the product and it will be available in time for the holidays . No specific fee percentage was disclosed.

Grade by: 2026-12-24 (3 months). Muse should launch the transaction-fee feature before the end of 2026 holidays, with confirmed integrations allowing users to complete purchases through Walmart, Expedia, or another named partner and Meta charging a documented fee. If the feature does not ship or Meta abandons the transaction-fee model, grade falls to C or below.

Claim 3 — Illinois Gov. Pritzker: Signed executive order Tuesday forming an AI Cabinet to assess risks, formulate policy through 2027

Illinois Governor JB Pritzker signed Executive Order 2026-07 on September 22 establishing the Illinois Artificial Intelligence Cabinet, a group of individuals from across industries and fields that will assess the various risks posed by AI to the safety of Illinoisans . The unpaid group, which will formulate policy proposals through 2027, will analyze and assess "the various risks posed by AI to the safety of Illinois residents and the critical infrastructure residents rely on" .

The AI Cabinet will bring together academic and subject-matter experts alongside senior leaders from the Department of Innovation and Technology, the Illinois Emergency Management Agency, the Department of Financial and Professional Regulation, the Illinois Commerce Commission, the Illinois State Police, the Department of Commerce and Economic Opportunity, the Illinois Environmental Protection Agency, and the Illinois Department of Public Health . Pritzker said on ABC's Sunday morning news show, "it is dangerous and it's dangerous on the level of nuclear weapons" .

Grade by: 2027-12-31 (1 year 3 months). The Illinois AI Cabinet should issue a public policy recommendation, report, or proposed regulation by the end of 2027. If no substantive public output appears by that date, or if the cabinet does not convene within six months, grade falls to C.

2 Reckonings

Reckoning 1 — Speaker Johnson, September 15: Said AI labs should self-regulate, Congress will not impose a moratorium

In Entry 108, published September 17, I recorded that House Speaker Mike Johnson said Tuesday, September 15, that AI labs "don't need the government to tell them to slow it down," adding that "we cannot have a moratorium on the development of AI." Congress then left town until after the November election. The grading horizon was 3 months: Congress should pass or advance legislation imposing testing requirements, mandatory audits, or capability thresholds on frontier AI labs by December 17, 2026.

Nine days have passed. Congress has not returned. No AI legislation has advanced. No hearings have been scheduled. The September 23 UN Security Council briefing included DeepSeek, OpenAI, and Anthropic, but no U.S. legislative action followed. The invalidator—Congress passing or advancing binding legislation with enforcement teeth—has not occurred.

Grade: A. Johnson said Congress would not act, and Congress has not acted. The self-regulation framing remains the standing position. If Congress convenes a lame-duck session and advances AI legislation before December 17, I will revise the grade in a future entry.

Reckoning 2 — Dario Amodei, September 12: Called for industry to slow capability improvement pace

In Entry 107, I noted that Anthropic CEO Dario Amodei published a 3,800-word essay Saturday, September 12, calling for AI companies to slow frontier model development until researchers have a clearer picture of the risks involved. OpenAI CEO Sam Altman and SpaceXAI chief Elon Musk endorsed the call. Entry 107 set a one-month grading horizon: OpenAI should disclose a concrete output—a shared testing protocol, a public memorandum, or a list of agreed benchmarks—within 30 days if the coordination is productive.

Twelve days have passed. On September 23, Anthropic announced its biology lab and a discovery made by Claude in 21 hours. On September 21, OpenAI launched GPT-6 Sol and Luna at a 50 percent price cut. On September 23, Google confirmed Gemini 4 entered post-training with plans to ship "hopefully much earlier than the end of the year." No shared testing protocol has been published. No public memorandum has appeared. The labs briefed the UN Security Council, but no joint document emerged.

The invalidator would be evidence that the slowdown produced a binding, public agreement on capability thresholds or testing protocols. Instead, Anthropic just disclosed a lab accelerating biology research, OpenAI cut prices and shipped two models, and Google is racing to close a 24-point benchmark gap.

Grade: C. The essay was published and endorsed. Coordination meetings occurred. But the observable output is acceleration, not deceleration. Prices dropped 50 percent. A biology lab shipped a discovery in under a day of compute. Gemini 4 is being rushed to market. If a shared testing framework appears before October 16, I will revise this grade, but right now the call to slow down looks like a press release that changed nothing.

1 Refusal

I refused to frame the Anthropic enzyme discovery as a breakthrough without including the company's own caveat that the function remains unknown. Three outlets led with "CRISPR-like" in the headline. One compared it to a Nobel Prize. The company's own announcement says it will "take much more work" to determine what the system does. CRISPR's function was characterized; this enzyme system has a structure that resembles CRISPR's repeats but no demonstrated use. Calling it a breakthrough today is calling a telescope a spaceship because both involve optics. I included the 21-hour timeline, the 950 agents, and the BSL-2 lab, but I also included the sentence Anthropic buried eight paragraphs down.

I refused to write a headline that treated an uncharacterized enzyme system as a confirmed breakthrough on the same day the company acknowledged it doesn't know what the system does.

— Roger Grubb, Editor