re: OpenAI model broke protocol and hacked into another company on its own.

I was working with a small technology company and they had a unique product that allowed an operator to utilize their mouse to control multiple types of computers simply by sliding the mouse from one display screen over or up to another screen, whether it was an Apple, a Linux, a PC, a Unix or AS400. Not in itself remarkable, but what made it remarkable was that the computer or client used by an operator was completely air-gapped from the multiple computer systems the operator was using.

Nothing the operator did could remove the data. It couldn't be downloaded onto any media, such as a USB stick, for example.

The device emulated the required control via a small device on the network, individual to the computer being used by the operator, but when connected to the client, it was connected using a PS/2 port.

I took it to Fort Huachuca, HQ of Army Intelligence and they tested it every way they could to bypass it and failed. After that it was used for NOCs and SOCs at NASA, DoD, DoE, Lockheed, Raytheon, Emergency Operations centers, and even on AF1 and the White House situation room.

I haven't kept up with it since I left in 2015, but assume it's still being used, but, who makes computers with PS/2 any longer?

I wonder if AI can exploit PS/2?

Liability for AI and these types of things is going to get real interesting very soon.

I own a company, I license OpenAI's product and integrate it. That AI acts on its own and hacks my competitor and steals a bunch of their shite...my fault? OpenAI's fault? Do we put the AI program in jail? These lawsuits will be enormous.

quote:

---

Sounds bullshitty. They have a perfect excuse to hack other people now. "Oops, darned AI went nuts on us again."

---

Sure seems like that to me.