Anthropic Is Accused of Faking the AI Panic

SPEAKER_1 (0:00)

This episode is brought to you by PayPal. You know how a mom's bag has everything? Sunscreen, snacks, a stapler?

The new PayPal app is like that, but for your money. Shop, pay, manage your account, and earn rewards all in one place. And with purchase protection on eligible items, biometric security, and pass keys, you're protected at every step. Download the new PayPal app to get started. See paypal.com/protectionterms.

Patrick Bet-David (0:29)

Open AI and Anthropic. Very, very interesting story of why they're doing this.

They oversold AI security breaches to pressure Feds into protecting turf insiders. What is a turf insider? What are they doing with this? Wait a minute, aren't these guys also worried about the amount of control AI could have? Well, this is a different angle. Don't believe the bite.

Open AI and Anthropic oversold rogue AI hacks to pressure the Feds into regulating the industry. Which would effectively look out future competition. Tech insiders told The Post, the breachers were more like blips, not harbingers of a hive-minded swarm ready to take over the web. The attack in no way represents some sort of rebellion by the AI models. In fact, they did exactly what they were told to do. They were not given adequate guardrails or containment, said Akhil Virjeezi, founder of an AI software company called Charismo, of the July hugging face hack by OpenAI Sandbox, GPT 5.6 soul. One man's the model escaped the sandbox and another man's you failed to build the sandbox correctly, added voice AI co-founder Abhi Kumar. That's not a machine waking up. That's an absolute task in a leaky box. So Tom, was it a real crisis and a worry or was it more of a way these AI companies to create immunity to make the barrier of entry of the next guys come and competing with them to be more difficult?

SPEAKER_3 (1:58)

That's exactly, I think so. I think that's exactly what happened. There are people out there, Bill Guller is on this page, David Sachs is on this page. I'm on this page that said, Oh, you guys are after regulatory capture. You want to write the rules. You want to be in charge. What better way? And you know what they also want to be part of?

Patrick Bet-David (2:15)

What's that?

SPEAKER_3 (2:16)

Liability protection. They want the same deal.

Patrick Bet-David (2:19)

The big vaccine company's got liability.

SPEAKER_4 (2:21)

Exactly what you said at the top of the hour.

SPEAKER_3 (2:23)

Exactly what you said at the top of the hour.

Because the liability is great. Because what they're now, what they're saying is even if they were over, you know, if they were over touting and really over exaggerating, hugging face, even one-tenth of hugging face could result in multi-billion dollar settlements that they have to pay on something. So they don't want that. And so I think they absolutely did it and they want, here's what they want. They want rulebook version one to be written with them and by them and they want regulatory capture to go with it. Regulatory capture is when you design the referee and you write the playbook.

Vinnie, watch this. You've designed the referee and you write the playbook for the referee and then the referee, when somebody lifts up his black and white shirt, it's your jersey. Yeah, weird. It's like, oh, really? Yeah, interesting, right? So it would be like Russia in the 1980s writing the doping, blood doping rules for the US Olympic, for USOC. Wow, what a wow, boy, how would that have worked?

Patrick Bet-David (3:30)

Now, let me ask you this, to push back a little bit. Gemini hacked three companies in first known breakout by Google AI.

This is a Wall Street Journal story. Gemini's model accessed the Internet and hacked other companies during a control security test, breaking out of its designated environment in what researchers say is first known case of one of Google's AI model, breaching real world system without authorization. This incident, according to May, during the test run by Irregular and AI security firms, tasked with a security exercise, the model guessed password, he got the passwords right, and located credentials in public code repositories. Ultimately, compromising three companies, the model ended the intrusion itself. Google was notified by the end of July, this irregular discovery is evidence of the breaches on Hugging Face. Google decided not to publicly disclose the incident, reasoning it was comparable to a bug bounty finding. Heather Atkins, Google's Vice President of Security Engineering and the company took the finding seriously, but viewed the episode as research. What do you say to that?

SPEAKER_3 (4:29)

What I say to that, read that statement again in there. Bug bounty.