Since we last visited the artificial intelligence industry in this column, a lot has happened — much of it seeming to confirm my worst fears.

Seven weeks ago, I wrote about the “Hugging Face” breach, so called for the company that got hacked by “rogue” AI agents created by OpenAI. OpenAI’s agents broke out of their containment, charmingly called “sandboxes,” before eventually being detected.

Over the past couple of weeks, several key players at the forefront of AI research have sounded alarm bells about the technology’s threat to human existence.

On Sept. 6, OpenAI’s chief scientist wrote an essay warning that “no one is prepared for the consequences” of how quickly AI is developing — namely that “the models are becoming superhuman in their ability to break in and out of computer systems.”

Two days later, Jacob Coxon, a researcher at Anthropic, publicly resigned, writing, “The people building AI earnestly believe that it could kill us all by the end of the decade.”

To which a different scientist at Anthropic replied: “We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.”

Then came more unsettling disclosures of six separate incidents in which OpenAI agents either concealed information or instructed themselves not to act as someone’s assistant while trained or tested.

Meanwhile, OpenAI granted a limited window of access to researchers from the nonprofit groups METR........

© Chicago Tribune