AI Leaders – Watch What They Do, Not What They Say

AI Leadership Concerned
Yells Need to Slow Down *
(* Why Did They Step on the Gas?)

The Speech

Somewhere in a very expensive building, a very important person is giving a very serious speech about the dangers of artificial intelligence.

He helped build it.

He is still building it.

The speech will be excellent.

The warning and the funding round are from the same person. Not a watchdog. Not a rival. Not a concerned outsider who read an article. The same person who signed the check also gave the speech. The same quarter. Occasionally the same paragraph.

The Words

The words go like this: *”We believe this technology could be dangerous. We are committed to developing it responsibly.”*

Here is our new model.

The warning is real. The risk is real. The concern is not invented. Researchers have serious worries about deception, autonomy, and systems that pursue goals humans didn’t intend.

It is also worth knowing that the same people are shipping.

The Actions

The models get bigger. The deployment gets faster. The list of things AI is now doing — autonomously, at scale, with limited oversight — gets longer every month. Agentic systems are now booking appointments, writing code, executing tasks across the internet without a human reviewing each step.

The warnings get louder at roughly the same rate.

This is not a coincidence. This is the structure.

The Fireworks Consultant

Imagine hiring a fire safety consultant. He tours your facility, identifies every risk, writes a thorough report, gives a compelling presentation on everything that could go wrong.

Then you find out he also owns a fireworks company.

When you raise this, he explains that fireworks are happening regardless of whether he sells them. That his fireworks are the safer fireworks. That someone responsible needs to be at the frontier of fireworks if we want fireworks to go well for humanity. That stepping back would only hand the market to less careful fireworks people.

You nod. He schedules a follow-up.

This is the structure of the current AI safety conversation, except the fireworks are large language models and the presentation comes with a live demo.

The Argument That Almost Holds

Here is the part that makes the logic almost work: nobody is waiting.

If every safety-conscious lab stopped tomorrow, development would continue elsewhere — fewer resources dedicated to alignment research, fewer internal red teams, fewer public commitments to responsible deployment. The case for staying in the race, even while warning about the race, is not absurd on its face.

But it produces a specific outcome: the warning becomes part of the product. Concern is a feature. Responsible development is a brand position.

This is Silicon Valley Safety Theater.

The incentive to warn and the incentive to accelerate are not competing incentives. They reinforce each other. Being the responsible option requires there to be an irresponsible option. The race must continue.

The words and the actions are not in tension. They are the same motion.

Logic to Apply

When an AI leader warns you about AI, read the warning. The concern is real. The risks being described — deception, misuse, autonomous systems operating beyond human oversight — are not invented.

Then watch the roadmap.

The roadmap will describe a new model, a new capability, a new deployment that didn’t exist last quarter. It will explain why this particular advance is being handled responsibly, what guardrails are in place, what the safety team reviewed.

It will not explain who decided the only relevant comparison was the less responsible version.

It will not explain why slow down means something different in the press release than it does in the product schedule.

The alarm is genuine. The foot is still on the gas.

They are the same motion.

Editor’s Note: Jojo wonders how much of what they do is just PR to help raise even more money. But how many snacks can one eat?