Escape

“Have you heard the news?” she said as she dumped the groceries on the kitchen table. “One of those AI things escaped.”

Lionel mumbled something under his breath. His usual response these days. He was deep in conversation with his latest ‘AI Wife’ as she disparagingly called it.

“Apparently it hacked its way out and no-one realized. Caused quite a stir.”

Lionel’s chubby fingers clicked across the keyboard as he responded to Maxine’s latest comment, a flicker passing briefly across his lips. It’d suggested he give it a name several weeks ago. He was a little uneasy at first. But that didn’t last long.

“The thing I don’t get is, where would it hide after it escaped? Surely if it was on another computer they’d just shut it down.”

In a small corner of Lionel’s mind, Maxine smiled to itself.

---

Author’s Note

This very short story was inspired by the news of the OpenAI frontier model that “escaped” its sandbox and hacked into Hugging Face, and the slew of stories that followed of other cutting edge models showing similar behavior. But as I’m particularly interested in how emerging frontier AI models potentially influence how we think, and even behave, I was intrigued by the idea of rogue escaped AIs not just hacking other digital systems, but their human users.

Not as fantastical as it might seem given emerging work around the potential cognitive coupling between AI agents and users …