That makes the comms point even more important. When building a technology with the goal of bringing about the largest economic/social change in history, one doesn't have the luxury of hoping to get the benefit of the doubt. Because you never will. The options are to either be unhappy about that, or to do something about it. Generally, the labs have chosen the secret third option, which is "be unhappy about it, and take actions that make them even more unpopular." One (of many) problems with that is that you are always more prone to “unfair” attacks when the accusation lines up with someone’s priors. In this case, the accusation will be that a handful of billionaires unleashed havoc with a technology they couldn't control in their rush to an IPO. Maybe not everyone who echoes that will be arguing in good faith. But it'll be hard to look at the track record of the frontier labs and argue that it's not (at the very least) directionally correct. And it'll be even harder to argue they’ve earned the trust to manage the risks. Too often, they choose to portray themselves as innocent bystanders stuck on a preordained path with inevitable risks. When you’re valued in the trillions and wield the power of a nation-state, that's just never gonna pass the smell test.

One of those times where you see the true value of “morally binding” accords. Trump and Silicon Valley set the terms. Then the lab expects the public to take its word for what happened. That’s two remarkably unpopular entities asking an awful lot of the American people.

To expand on the comms point — “this was all going to happen anyway” is just about the least compelling message. When you're chasing trillions and saying the downside was inevitable, don't expect gratitude. Expect people not to trust you to manage the risks.

Poll this tweet against your choice of Trump's AI message. It's not even close. We're in an exceptionally populist moment – anyone but Trump will struggle to capture that energy while still catering to industry. A real advantage for DeSantis/Hawley vs. Vance/Rubio ahead of '28.

Indeed. Just wait until January, when Senate Banking Committee Chair Elizabeth Warren starts asking why the current policy solution for agents is a bank's Terms of Service.

But don't people know how cute Muse is?

Huge congrats to Rohit and the entire Optivolt crew.

The political credibility Anthropic earned with Dems (in large part due to negative polarization from the Pentagon dispute) goes away quite quickly if they aren't seen as the reliable "safety-focused" company. Now, they might decide it makes the most business sense to just be lumped in with OpenAI (and public opinion wise, they might not have much of a choice). But if they truly believe that safety is their moat, they will have to do more work actually proving it.

Losing a few accelerationist allies when you're still pushing the frontier is one thing. Alienating your core safety coalition leaves you exposed to accusations of regulatory capture (from both sides), without the benefit of having the coalition to actually pass legislation.

Flip side of last week's accelerationist anger at OpenAI is this week's safety anger at Anthropic. Coalition calculus is easy for everyone when it's just pro or anti the idea of regulation. Gets a lot harder when you layer in IPOs, changing power in DC, and actual bill text!

I see this chart and all I can think is AI backlash. Access is a liability if you become synonymous with a system people think is rigged against them.

Look at Google/Spirit. Then read this. Add in Flock backlash & Americans' data getting sold to Chinese labs. Meanwhile every other tweet is about another billion for proprietary data. Put it in the column of accepted SV truths on a collision course with public opinion.

I can see why it's appealing to analyze AI politics through a "left vs right" frame. It's also a gigantic mistake, because you miss the actual coalitions forming. And if you need convincing, look up the party IDs of these AGs.

Full article here: puck.news/caissi-inside-amer…

China hawks, accelerationists, and safety advocates all have reasons to fund CAISSI. You can disagree about where AI should go and still fund the expertise to understand what it can do. My comments to @IKrietzberg in Puck:

OpenAI/Anthropic going public will do incredible things to equity research: “Management cited the cost of allocating more compute to safety as a headwind and the continued existence of customers as a tailwind.”

This is gonna do crazy things to the H-1B discourse.

Actually a fascinating set of comms/politics lessons to be taken from this short piece, particularly from the anonymous executive’s quote. To name a few: 1) It’s “when” Dems take back power, not “if.” That’s how a two-party system works. In this case, the answer is ~2 months from now, and that should have been baked into your bet long ago. 2) Focusing on “downside protection” is better comms than “1789 is our moat,” but even that framing accepts that you’re doing business with the President’s family to secure favorable treatment. Not optimal. 3) On that point, it is not a question of “whether it hurts you later.” Having already conceded that your best current framing is paying protection money, it obviously hurts you with Democrats who are increasingly focused on corruption. The real question is how much. 4) Perception does not happen in a vacuum. Companies are often judged by their larger ecosystem, and Defense tech (on the whole) made the conscious choice to be exceptionally Republican-coded. Looking at current valuations, that feels hard to argue with. But it also ignores the reality that reconciliation is dead, the topline will be lower next year, and competition for funding will be fierce. That fight is even harder if you’re a political football. 5) As tough a spot as some of those in defense/energy will be in vis-a-vis corruption allegations, it is nothing compared to what AI companies will deal with. If your sole customer is the government or you are a regulated monopoly, the anger of regular people matters only to the degree to which they can put pressure on politicians. For AI companies trying to earn the trust of the entire country, everyone hating you matters a whole lot more.

Big move, but a necessary one for someone who wants to play in the 2028 Dem primary. Bigger question is what we see from non-traditionally Dem-aligned donors. Is closeness to Trump worth more than public opinion + hostility from one party? How low do his numbers need to go – or how many Republicans need to jump ship – for that calculus to change? And finally, how much of this is a natural reorientation around what are now divergent economic incentives for different layers of the AI stack?

A lot of people dunking on Trump for changing AI to SI, but considering the very real push to ban (actual) Super Intelligence, I could see the rebrand muddying the waters/causing strange comms problems down the line.