Anthropic confirms real world AI weaponization

SPEAKER_1 (0:00)

This episode is brought to you by ChatGPT.

Hey, it's Bill Simmons from the Bill Simmons Podcast. Have you guys heard about ChatGPT work? It's the new way to use ChatGPT for bigger multi-step projects. And when you need more than just answers, you have ChatGPT work, access to your apps and files, and it can create real work documents like spreadsheets, slides and structured reports. Get started at chatgpt.com by selecting work mode available on Plus and Pro plans.

SPEAKER_2 (0:31)

I see you. Avatar Fire and Ash is now streaming on Disney Plus.

It's the film critics are calling the best Avatar yet.

SPEAKER_4 (0:39)

Go, go, go, go.

SPEAKER_2 (0:40)

A true epic and completely jaw dropping.

SPEAKER_4 (0:42)

This is the only pure thing in this world.

SPEAKER_2 (0:45)

Return to Pandora on Disney Plus.

SPEAKER_1 (0:47)

It will be an adventure for the whole fam.

SPEAKER_2 (0:49)

And watch the Oscar-winning phenomenon at home. This is sick.

Avatar Fire and Ash now streaming on Disney Plus. Rated PG-13.

SPEAKER_4 (1:00)

Anthropic released a 154-page threat intelligence report, detailing exactly how foreign security services, scientists and opportunistic hackers have circumvented the safeguards on Claude AI models. We are looking at documented attempts to use these models for biological weapon development, advanced missile design and automated cyber attacks.

SPEAKER_3 (1:21)

Yeah, that document represents a very specific threshold being crossed in the artificial intelligence industry. I mean, it is the first known admission from a major developer that biological weapons development actively overlaps with AI misuse.

SPEAKER_4 (1:36)

Right.

SPEAKER_3 (1:37)

Before this release, you would really only read about that kind of overlap in academic research papers or speculative government briefings. International organizations have been outlining these scenarios for quite some time, but the companies actually building the models have maintained a pretty clear distance from that public reality. Until now.

Exactly. Now we have a private company offering concrete evidence that their platform was utilized for exactly this purpose.

SPEAKER_4 (2:00)

It takes the whole conversation out of the theoretical realm and places it firmly into applied science.

We are no longer discussing what an adversary might do if they gained access to a highly capable system. We are looking at server logs that show exactly what they attempted to do.

SPEAKER_3 (2:16)

Yeah, the theory is over.

SPEAKER_4 (2:17)

And the report provides a very specific biological case study to illustrate how this automation is being applied in practice. It involves gain of function research on the Chikungunya virus.

SPEAKER_3 (2:28)

Which is a mosquito-borne pathogen, right? Yeah. It causes severe joint pain and fever.

SPEAKER_4 (2:32)

Right. And the actors involved in this case requested Claude to assist in writing a grant application for scientific funding.

The specific focus of that grant was the virus' transmissibility and its immune evasion properties.

SPEAKER_3 (2:45)

I have to question the utility of using an advanced artificial intelligence just to write a grant application.

SPEAKER_4 (2:52)

It sounds a bit mundane.

SPEAKER_3 (2:53)

It does.

On the surface, it sounds incredibly mundane. When you hear about an AI being used for bioweapons, you expect it to be generating, I don't know, a novel chemical recipe or a highly lethal genetic sequence. Instead, we are talking about administrative paperwork. Yeah. It just seems like a low-level application of a highly sophisticated tool.

SPEAKER_4 (3:16)

Well, the paperwork is the actual weaponization of the process. If you look at the real-world consequences of scientific funding, the administrative layer is the primary bottleneck for dangerous research. Okay.

SPEAKER_3 (3:28)

I see where you're going.

SPEAKER_4 (3:29)

AI is not just designing the science in a vacuum. It is actively accelerating the bureaucratic and funding mechanisms required to get that research off the ground in the first place. Getting a grant approved requires synthesizing massive amounts of literature.

SPEAKER_3 (3:42)

You have to identify precise gaps in current virology.

SPEAKER_4 (3:46)

And formulate a hypothesis that actually passes an institutional review board. That process usually takes human scientists months of dedicated, highly specialized labor.

SPEAKER_3 (3:56)

That administrative friction serves as an unintentional safeguard for the entire scientific community. The barrier to entry for complex, highly dangerous research is being lowered not just on the laboratory bench, but in the paperwork that enables the laboratory to exist at all.

You really have to consider what it means when the synthesis of decades of virology research specifically regarding immune evasion can be compiled, formatted, and optimized for a funding committee in a matter of seconds.

SPEAKER_4 (4:26)

It is essentially doing the labor of a seasoned principal investigator.

SPEAKER_3 (4:29)

Yes. It is organizing the entire theoretical framework necessary to justify enhancing a pathogen.

SPEAKER_4 (4:37)

It removes the necessity for a certain caliber of human expertise in the planning phase. If a bad actor wants to explore how to make a virus more transmissible, they historically needed a team of highly educated specialists just to conceptualize the project.