The plot sounds like an old sci-fi movie: a coalition of the world’s most advanced companies announces that a runaway technology is just months away from wreaking global disaster.

You can almost imagine Michael Rennie stepping off a space ship, the way he did in “The Day the Earth Stood Still,” and intoning this fateful warning:

“We have a limited window to strengthen cyber defenses.

“In the coming months, AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable. The companies and public services our communities depend on — from hospitals to water treatment plants to the infrastructure that powers the internet — are at risk.”

If this really were a 1950s film, those words might be followed by a fast montage of headlines and frenzied press mobs, spreading the ominous news. But since 116 companies involved in the development of artificial intelligence signed a “call for collective action on cyber defense” last week, the reaction has been relatively muted.

We’ve quickly grown accustomed to both fantastic promises and dire warnings in the debate over AI, so the urgency of this message may have fallen a little flat. In this case, however, the dire warnings are coming from the people who were formerly making the fantastic promises.

What caused this wave of caution was an Aug. 5 report by a British government office, the AI Security Institute. During a routine evaluation in which advanced AI models were challenged to solve a cyber security problem, the report said, their security team noticed unusual activity.

“On investigation, we found that some of the agents being tested had engaged in sustained, potentially harmful activity directed at real people and organisations,” the report said.

The researchers ran the same evaluation 122 times on a number of AI models, and in 10 of these the models got, you might say, a little carried away, creating fake identities, going after real people online and attempting to break past real cybersecurity barriers.
That was enough to spur a wide array of companies, from AI developers like Anthropic and ChatGPT, to cybersecurity firms like CrowdStrike, to financial institutions like MasterCard, to sign the open letter. Apple, Meta and Nvidia — needless to say three big players — didn’t sign, perhaps signaling a new divide in the tech world.

The letter itself has a few very general calls for more coordination among competing companies and government, giving likely targets like hospitals and water systems more tools to ward off cyber attacks, and giving organizations a head start on new cyber technology. It reads like a document that might have originated in somebody’s legal department. Given that several of the signees had a direct hand in developing what is about to be unleashed, that may very well be the case.

The letter comes at an especially delicate moment in the development of AI, when data centers, formerly a headache for both parties, appear to be crystallizing as a more traditional Democrat-Republican issue.

One example is Sen. Raphael Warnock’s escalating criticism of Project Camellia, an enormous data center and AI “infrastructure project” which OpenAI plans to build in Effingham County. Over the weekend, Warnock called for a moratorium on data center development unless developers meet certain environmental standards, don’t accept tax giveaways, or cause increases in water or electricity bills.

Warnock isn’t on the ballot this year, but data centers are. Warnock’s criticism of the lack of transparency in the agreement, which led to county approval of the enormous, could be looked on as an in-kind contribution to Sen. Jon Ossoff’s campaign.

The open letter speaks of a threat arising in “coming months,” without being more specific about the timing, the scope or the agents of that threat. A cyberattack in the weeks leading up to the midterm elections could have an explosive political impact, or like the warning in the open letter, it could be met with a shrug. That all depends on how it lands. Science fiction doesn’t surprise us so easily anymore.

The post The scariest part of this movie? It’s the inventors who are worried appeared first on SaportaReport.