On Sunday, Anthropic CEO and cofounder Dario Amodei expressed concern that within the next year a swarm of AI agents could take over the entire internet. “The scale of damage would continue to increase from there if AI becomes more powerful without the necessary guardrails,” he wrote just days after one of his own staffers quit over concerns that America’s premier AI labs are racing to create something that could wipe out humanity.
That sounds dire. But cybersecurity experts don't believe an AI will take control of the internet.
Ciaran Martin, former head of the U.K. government’s National Cybersecurity Centre, wrote that Amodei’s warning was “very sloppy, easily dismantled,” and dismissed the risk of “digital cyber apocalypse.”
“This is not a credible warning,” Martin wrote. “It assumes no monitoring of systems, no anti-virus, no DDoS protection, no network segmentation, no incident management, no nothing of any kind of cybersecurity on the global Internet of the type that has developed over the last 30 years.”
“For a claim of this magnitude, there is neither evidence for the contention nor a credible account of a path to this outcome.”
Amodei’s warning didn’t come out of the blue. It followed the now-infamous breach at AI company Hugging Face, when rogue OpenAI agents broke free of their testing environment to break into a number of companies. It later emerged Anthropic’s own models as well as those controlled by Meta carried out similar attacks. The coordination and persistence of the offending AI agents caused valid concerns. The ability for agents to gain internet access and create secret communications channels was especially eyecatching.
But Martin argued a wider catastrophe isn’t likely. “The cyber risks associated with all this seem manageable,” Martin added. Marcus Hutchins, a cyber threat researcher and ex-cybercriminal who once helped stop the spread of the WannaCry malware, wrote that Amodei was “fixated on imaginary cyberattacks that are literally impossible."
AI agents remain adept at exploiting known weaknesses, however. As Anthropic noted in a post in August, its AI agents used “basic techniques, such as exploiting weak passwords and unauthenticated endpoints,” to carry out their attacks. If the targets had done even rudimentary work to protect their systems, they wouldn’t have been hacked.
Got a tip on surveillance or cybercrime? Get me on Signal at +1 929-512-7964. |
Komentar
Posting Komentar