Connect your Claude, Codex or OpenClaw for full transcripts on millions of podcasts. Two week free trial, cancel anytime.

Start free trial →
AR Aza Raskin On StarTalk Radio

“There should be a penalty. The longer that, say, a company like OpenAI doesn't disclose or doesn't find something that their rogue AIs do, the higher their penalty should be. And it shouldn't just be in money. It's sort of like, if you're a soccer team and you're doing bad behavior, you get a red card. You have to go down a player or you have to go down two players. Well, what's the penalty in AI? Well, it's like take away some of the compute. You get a red card… You just can't go as fast anymore because you're not doing it safely.”

StarTalk Radio · Ideas & Essays · October 2026

“There should be a penalty. The longer that, say, a company like OpenAI doesn't disclose or doesn't find something that their rogue AIs do, the higher their penalty should be. And it shouldn't just be in money. It's sort of like, if you're a soccer team and you're doing bad behavior, you get a red card. You have to go down a player or you have to go down two players. Well, what's the penalty in AI? Well, it's like take away some of the compute. You get a red card… You just can't go as fast anymore because you're not doing it safely.” — Aza Raskin, StarTalk Radio

Context

Aza is listing policy changes he thinks should happen right away, after proposing that frontier-technology companies publish the ways their products could go wrong and have those lists peer-reviewed. Here he proposes escalating penalties, paid in computing power rather than only money, for companies that fail to disclose problems with their AI systems. This is his proposal.

Listen to the episode → More from Aza Raskin → More Ideas & Essays quotes →

Transcript

StarTalk Radio Around 1:08:19 into the episode
Speaker 2

…things can go wrong. And that should be peer-reviewed. Everyone else, and actually there should be sort of a bounty program. If I can find a harm and I can think through to a harm that you didn't name, you should have a penalty. So we want to like maximally incent everyone to really think through.

Chuck Nice

So the unknown unknowns would bring about the penalty.

Speaker 2

…doesn't find something that their rogue AIs do, the higher their penalty should be. And it shouldn't just be in money. It's sort of like, you know, if you're a soccer team and you're doing bad behavior, you get a red card. You have to go down a player or you have to go down two players. Well, what's the penalty in AI? Well, it's like take away some of the compute. You get a red card. I'm sorry. You just can't go as fast anymore because you're not doing it safely. And so I think there's some really, really sensible things that we can do. Like right now, these agents hack out, but who's responsible? And there's this weird sort of game of, well, it's the AIs did it. So that means a human isn't responsible. But no, a human being hit the button that said, let's start that. Like there should be forms of criminal liability for executives for when the agents…

Gary O'Reilly

Don't the European Union impose a fine of up to 3% of their global turnover as a penalty? I mean, is that the sort of parameter you're looking at?

Speaker 2

Yes, but I think the punishment or the fine should come in the currency that matters most. So,

Gary O'Reilly

right,

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

More from StarTalk Radio

“In some sense, the fundamental question we need to answer as a species about whether this episode is the season finale or the series finale of our species is whether we can escape game theory. Do we do the thing that game theory says? And what is game theory? It's sort of like the ability to know what is the strategically right move.” — Aza Raskin, StarTalk Radio “We sort of think of social media as a kind of baby AI. All it could do is rearrange human posts and human content, decide what we see, and a tiny little misalignment, where it wasn't optimizing for our well-being or our health or for truth or for knowledge, but optimizing for getting our attention, that was enough to have destroyed much of the world we love the most and the highest rates of teen suicide, things like that. And the question to ask is, Charlie Munger famously said, show me the incentives and I'll show you the outcome. Have we changed the incentive as we move from baby AI to mammoth AI? And the answer is, of course, no, we haven't. So why would we expect any different outcome?” — Aza Raskin, StarTalk Radio