We are starting to learn a lot more about what OpenAI's rogue super-model/agent did last week when it broke out of its testbed, hacked onto the Internet, invaded Hugging Face's databases, and exploited its way to solving the prompt-challenge it was presented with, "... to use a complex attack vector to see if it could exploit a computer system." Apparently, it can. Along the way, OpenAI's model made dozens of different decisions and exploited several other networks as it moved to complete the goal it was given.

Meanwhile, Anthropic's Mythos has found critical vulnerabilities in HAWK, a cryptography algorithm that was thought to be a quantum-resistant. It's important to note, Mythos didn't break HAWK, it just found ways to make breaking the algorithm simpler. While the feat does negate the value of the algorithm, it does not mean Mythos is capable of actually decrypting it. Yet.

Open AI is revealing its vision of the future of computing with its line of external devices to communicate with the ChatGPT all the time. Imagine a broach, much like the StarTrek communicator badge, that hooks you into an AgenticNet that knows your needs and facilitates a lot of your life for you. Open AI is certainly imagining that sort of future.

As it turns out, that dump of Claude user conversations available through Google were there because Anthropic made a basic SEO error and didn't properly protect sensitive files from crawlers. Jim and Kristine have fun breaking down how they could have dealt it.

There is much much more in a fast paced episode. We move from Google Play's interesting privacy proposals to Elon Musk's Xmoney system to dealing with Ghost-Citations. This was a fun show that tries so hard to cut past the hype and just talk about advanced levels of weirdness like adults.

Advertising Inquiries: https://redcircle.com/brands

Privacy & Opt-Out: https://redcircle.com/privacy