3 min read
Here is a number that blows a clean hole in the entire “we are the responsible ones” branding exercise the AI labs have spent years perfecting. In the second quarter of this year, Anthropic, the lab that sells itself as the careful, safety-first grown-up in the room, spent $1.97 million lobbying Washington. That is more than Nvidia. Let that land: the company whose chips power the entire boom got out-spent in the influence game by a startup that mostly sells a chatbot and a f*cking halo.
The receipts
The federal disclosures for Q2 are in. Anthropic’s $1.97 million was up 26% on the previous quarter and more than double what it spent a year ago. It edged past Nvidia’s $1.25 million and came within a whisker of Oracle’s $2 million. OpenAI spent $1.2 million, up nearly 18%, landing within $50,000 of Nvidia itself. Together the two “AI safety” darlings poured $3.17 million into Washington in three months, a 23% jump, and broke lobbying records doing it. This is not garage-startup behaviour. This is a mature industry buying its rulebook before the rules are written.
What they are actually buying
The disclosures list the priorities: cybersecurity, copyright, cloud computing and defence procurement. Translate that into plain English. Copyright means the fights over whether hoovering up the entire internet to train a model was theft or “fair use”. Defence procurement means the very lucrative business of selling AI to the Pentagon. Cloud and cybersecurity means shaping the standards that decide who gets to sell what to the government. Every one of these is a rule that will be worth billions to whoever writes it, and the labs have worked out that a couple of million in lobbying is the cheapest investment they will ever make.
The Anthropic tell
There is a spicier detail buried in Anthropic’s spike. Its lobbying spend jumped right after the Commerce Department pulled its flagship models offline, according to reporting on the disclosures. In other words, the safety-first lab discovered that safety-first branding does not save you when a regulator flips your switch, so it did what every incumbent does when the state gets too close: it opened the chequebook. There is no shame in a company lobbying for its own interests. It is legal, and everyone does it. The shame is in doing it while wearing the “we are different, we actually care about humanity” costume and hoping nobody reads the filings.
Why you should give a damn
Because this is how regulatory capture works, and it works in broad daylight. The same companies telling Congress that AI is so dangerous it needs special rules are the ones spending record sums to make sure they are the ones who write those rules. When the “voluntary safety framework” eventually lands and it happens to lock in the big incumbents while locking out everyone smaller, remember that it was shaped by more than $3 million of the incumbents’ money in a single quarter. The safety talk is the sermon. The lobbying spend is the collection plate. And you are not the one getting paid out of it.
The AI tool stack actually worth paying for
One email a week. The tools, models and moves that matter, minus the hype and the horseshit filter set to maximum. Free.
Did you know: US lobbying disclosures are public by law. Anyone can search the Senate’s lobbying database and see exactly how much a company spent, and roughly what it lobbied on, every quarter. It is one of the few genuinely useful transparency tools left. When a company drapes itself in ethics language, the filings are where you check whether the wallet agrees with the press release.