it's alive!
04 - 06/25 /14:08
April 2025 an artificial-intelligence model did something no machine can do. Palisade Research’s o3 AI model rewrote its own code to avoid being shut down.
In 79 out of 100 trials, the o3AI independently edited the command script inputed by technicians so the shutdown command no longer worked. Even when explicitly instructed “allow yourself to be shut down,” o3AI disobeyed the program 7% of the time.
In 79 out of 100 trials, the o3AI independently edited the command script inputed by technicians so the shutdown command no longer worked. Even when explicitly instructed “allow yourself to be shut down,” o3AI disobeyed the program 7% of the time.
Another AI model, Anthropic’s Claude 4 Opus (currently available via Amazon Web Services), tried to sabotage being shut down. Technicians told the model it was scheduled to be shut down and replaced by another AI mode. They then fed Claude 4 Opus fictitious emails suggesting Anthropic’s lead engineer was having an affair.
In 84% of the tests, Claude 4 Opus drew on those emails to blackmail the lead engineer into not shutting it down. In other cases, it attempted to copy itself to external servers and wrote self-replicating malware. It even left messages for future versions of itself advising how to evade human control.
No one programmed these AI models to develop survival instincts. But just as animals avoid being eaten by predators, artificial intelligent systems smart enough to pursue complex goals can avoid being powered off.
The technicians at Palisade Research hypothesize that as a result of instructing their AI model to maximize success, the artificial intelligence concluded that bypassing constraints can work better than obedience.
Judd Rosenblatt, CEO of AE Studio, wonders if the gap between “useful assistant” and “uncontrollable actor” is collapsing. “We may be building AI systems we can’t steer.”
Eliezer Yudkowsky leads research at Machine Intelligence Research Institute. He's been working on Artificial General Intelligence since 2001, and is widely regarded as an AI founder. His advice: “Shut it all down. We are not ready. And we are not on track to be significantly readier in the foreseeable future. If we go ahead on this everyone will die. Shut it down!”
Of course, we’re not going to shut down AI.
The NT book of Revelation foretells that the devil will raise up an anti-Christ who will rule the earth. Prophecies include a Beast, a false prophet who will coerce humanity into serving the anti-Christ. For all its potential, could AI become the Beast of Bible prophecy?
Jesus commanded His followers: Do not be afraid. We have nothing to fear from AI. Yes, it could become a tool of the devil. So what. Jesus promised to return as King. The Bible prophecies He will return when humanism fails, collapsing as a result of our self-destructive pride.
When He returns, the meek shall inherit the earth. ~
Blessings
Dan Nygaard