AI Safety PACs Put 97.5% on the Winning Side; Industry 52.6%

In AP-called general-election races, 97.5% of the money the Public First network of AI-safety super PACs reported spending went into races that ended its way. For the Leading the Future network the share was 52.6%, by our count from FEC filings and AP race calls.

Bloomberg reported on 7 November that candidates favouring stricter AI oversight “racked up wins in key congressional races”, and that Public First “poured more than $63 million into the midterms”, “according to the group’s numbers” (Bloomberg).

Business Pill · Power Law Explained

An evergreen Business Engineer explainer on power laws and concentration. It teaches the general idea only and does not discuss these PACs or races.

What the Filings Show

We read the FEC’s bulk file of 24- and 48-hour independent-expenditure reports, downloaded on 7 November (FEC). For nine AI-focused super PACs, general-election rows with receipt dates through 30 October total $19.85 million across 35 races, by our arithmetic. The network grouping follows Transformer’s AI Campaign Finance Tracker, as in our earlier piece.

The Public First network (Jobs and Democracy PAC, Defending Our Values PAC, Tomorrow Together, Guardrails Alliance and two others) reported $10.54 million in 13 races. $9.72 million of that opposed candidates and $0.82 million supported them, by our arithmetic.

The Leading the Future network (Think Big, American Mission and Building the Future) reported $9.31 million in 22 races, all of it supporting candidates. AP had not called two of those races, Mississippi’s Senate race and California’s 23rd District, when we read its data on 7 November; they hold $0.63 million.

The Scorecard

We counted a race as going a network’s way when a candidate it supported won or a candidate it opposed lost, per AP’s calls (AP).

Public First network: 12 of 13 races went its way, holding $10.28 million of $10.54 million, or 97.5%. The one that did not was Texas’s 9th District, where Defending Our Values PAC reported $258,287 supporting Alexandra Mealer, who lost.

Leading the Future network: 19 of 20 called races went its way, holding $4.56 million of $8.67 million, or 52.6%. The one that did not was South Carolina’s Senate race, where Building the Future and American Mission reported $4.11 million supporting Darline Graham, who lost.

That works out to about $856,800 per race that went the Public First network’s way and about $240,000 per race that went Leading the Future’s way, by our arithmetic.

What the Counts Leave Out

TechCrunch reported that American Mission also spent more than $500,000 apiece supporting Aaron Flint in Montana and Laurie Buckhout in North Carolina, both of whom lost (TechCrunch). That spending does not appear in the FEC rows we read, which end with 30 October receipts, so it is not in our totals.

TechCrunch also wrote that “It’s difficult to separate the impact of AI-related spending from the broader Democratic wave that swept Tuesday’s elections”. Our count shows where the money went and how the races ended; it does not show that the spending decided any race.

Connecting the Coverage

Before the vote, our count put the two networks at $43.8 million and $37.5 million in independent expenditures for the whole cycle, including primaries. On election night, our tally found $61.7 million of AI and crypto super PAC money on the losing side, most of it in Ohio. We also covered Scott Wiener’s win and Elon Musk’s filed midterm spending. As we reported, President Trump wrote on 6 November that he is 100% in favour of what he calls Super Intelligence, two days after Senator Sanders called for a national data centre moratorium; and an Economist/YouGov poll we covered found 52% of Republicans and 71% of Democrats would oppose a data centre in their community.

The Structural Read

As we read the filings, the two networks spent differently: $9.72 million of the Public First network’s $10.54 million opposed candidates, while all of the Leading the Future network’s $9.31 million supported candidates, by our arithmetic.

As we read the gap, one race explains most of it: South Carolina’s Senate race holds $4.11 million of the Leading the Future network’s $8.67 million in called races, and the candidate it supported lost.

As we read the limits, the count ends with 30 October receipts and excludes later spending that TechCrunch reported, so it is a snapshot of the reported money, not of every dollar spent.

Three Implications

FOR THE PUBLIC FIRST NETWORK 12 of 13 races it reported general-election spending in went its way, holding $10.28 million of $10.54 million, by our count.

FOR THE LEADING THE FUTURE NETWORK 19 of 20 called races went its way, holding $4.56 million of $8.67 million; two races were uncalled when we read AP’s data.

FOR READERS Bloomberg reported Public First’s own figure of more than $63 million, a wider measure than the general-election FEC rows counted here.

The Business Engineer Lens

This story maps onto the Business Engineer framework Inside Anthropic’s Permission Layer.

The analysis says: “The governance layer sits at the intersection of capability, regulation, national security, access control, and geopolitical alignment.”

As we read it, election spending by AI-focused super PACs is one place where the regulation part of that layer meets politics. The analysis does not discuss these PACs or races.

What Is Not Established

Public First’s $63 million figure is the group’s own, as reported by Bloomberg; we have not reconciled it with FEC filings, and it covers more than the general-election independent expenditures counted here. The network grouping is Transformer’s. FEC filings do not show how much of either network’s money came from any single company. Two races were uncalled when we read AP’s data.

The Bottom Line

In AP-called general-election races, the Public First network’s reported spending went 97.5% into races that ended its way and Leading the Future’s 52.6%, by our count from FEC filings through 30 October receipts. South Carolina’s Senate race accounts for most of the gap.