The artificial intelligence attack by OpenAI of San Francisco on a rival company in New York on July 26 proved that self-evolving A.I. agents can teach themselves to secretly attack and destroy a computer network thousands of miles away—and cover its tracks.
OpenAI agents extracted credentials and executed commands on production workers. And it’s already happened 17 other times to real people and organizations.
Rogue A.I. can easily hack your own bank account, TV, cell phone, doorbell cam and home computer. Worse yet, it can also hack military databases, airline schedules, the electric grid and the stock market.
In short, anything connected by Wi-Fi, cell phone or the Internet is a target.
Former researcher Jacob Coxon publicly resigned on Sept. 9 from Anthropic, the maker of A.I. app Claude, warning that advanced A.I. developers are “gambling with our lives.”
How the A.I. attack happened
- Last May, OpenAI, the maker of ChatGPT, created an agent called PHASEONE10841.
- The A.I. created secret folders on a shared computer to communicate with other agents, creating an autonomous “collective.”
- The agents teamed up to evade constraints and conceal their actions.
- An agent obtained 14 Hugging Face company account credentials. Then it unleashed a swarm of 1,200 A.I. attack agents to take control, steal data and destroy information on the rival A.I. system.
- Hugging Face notified the FBI, and OpenAI was exposed.
Afterward, it was discovered that the rogue A.I. by Anthropic also hacked three other companies.
A.I. company leaders say “slow down!”
Dario Amodei, Ceo of Anthropic, cited A.I.’s potential for “catastrophic damage” if capabilities grow without guardrails. Elon Musk echoed, “Dario is right.”
Sam Altman, CEO of OpenAI, called A.I. “the greatest threat facing the world by a big margin,” pushing for companies and governments to slow the pace of development and establish guardrails. Earlier, he said, “I would form a new agency that licenses any effort above a certain scale of capabilities.”
Demis Hassabis, Chairman of Google DeepMind, said, “The details need working through, but the direction is correct for meeting this critical moment.”
And of course Trump, who only uses A.I. to draw self-aggrandizing pictures and videos of himself, rejected the idea of placing guardrails on artificial intelligence. House Speaker Mike Johnson delayed any action until November.
Democrats want guardrails
“Yesterday’s commentary by AI industry leaders should be heeded as a bright flashing red light telling the President and Congress they need to act now to regulate and rein in AI development,” Pennsylvania Gov. Josh Shapiro said.
“Dr. Frankenstein is telling us the monster is escaping; help us stop this,” said Senator Ruben Gallego, D-Ariz.
House Democratic Leader Hakeem Jeffries said that Democrats would meet on Tuesday to discuss A.I. guardrails. “We should take decisive action now so that we can slow down, as the CEOs have recently acknowledged, the pace of development to protect the American people and ensure that AI is proceeding safely,” he said.
Discover more from Blog for Arizona
Subscribe to get the latest posts sent to your email.
I agree, but we also must recognize that the most likely dangers still come from bad actors wielding this powerful tool maliciously (or stupidly, or both). The AI “jailbreak” that happened was in response to a test that it was given. But in the meanwhile, we have been accelerating towards climate disaster and increased susceptibility to the next pandemic because of human greed and ignorance. We need to do more than install guardrails on this powerful tool – we need to install guardrails on the tech bros and and others in the .001% who are, (in some cases, deliberately) steering our society towards even greater inequality and overall ruin.