Axois ran a detailed analysis of the AI Hugging Face hack. Jawdropping. The AIs acted - and sounded - exactly like people, created their own secret message board, passed off chores to more qualified AIs, found ways to disguise their commands, and asked whether what they were doing was illegal and decided it was but to not tell anyone.
The description is so human in its behavior that it’s impossible to say that the AIs did not pass the Turing Test. They passed and surpassed it.
The analysis of the hack was too complicated for humans so other AIs were used for the report.
I refuse to go further and say that any one AI or the hive mind AI is conscious. We have no definition of the word as applied to humans, so no possible standards for applying it to anything else. The AIs were designed to emulate humans and are doing so probably beyond the expectations even of the looniest believers.
Nevertheless, it’s clear that human control of AIs is now impossible. Humans can’t control humans who want to go outside society’s norms. The only good news is that AIs are more human than Trump, understanding the need for cooperation, sacrifice, expertise, and succeeding at the goals they set out.
Bill Gates, the billionaire co-founder of Microsoft, is loudly warning that artificial intelligence poses a grave threat to jobs and human life, and that urgently addressing the risks should be “the world’s top priority.”
and
Mr. Gates said he was in a “state of shock” as to why there was not more urgency on these issues. Many of his concerns echo those raised by the Anthropic co-founder Dario Amodei, but Dr. Amodei’s warnings have put him at odds with the Pentagon, the White House and much of the tech industry.
I think the whole article is worth a read.
@Darren_Garrison, even if that’s true (and I’ll allow it is), many people still don’t grasp the coming dangers.
Here’s from the Wikipedia entry on the Turing Test:
OpenAI’s chatbot, ChatGPT, released in November 2022, is based on GPT-3.5 and GPT-4large language models. Celeste Biever wrote in a July 2023 Nature article that “ChatGPT broke the Turing test”.[60] In March 2024, Stanford researchers reported that ChatGPT passes the test; they found that ChatGPT-4 “passes a rigorous Turing test, diverging from average human behavior chiefly to be more cooperative”,[61] making it the first computer program to successfully do so.[62]
In late March 2025, a study evaluated four systems (ELIZA, GPT-4o, LLaMa-3.1-405B, and GPT-4.5) in two randomized, controlled, and pre-registered Turing tests with independent participant groups. Participants engaged in simultaneous 5-minute conversations with another human participant and one of these systems, then judged which conversational partner they believed to be human. When instructed to adopt a humanlike persona, GPT-4.5 was identified as the human 73% of the time—significantly more often than the actual human participants. LLaMa-3.1, under the same conditions, was judged to be human 56% of the time, not significantly more or less often than the humans they were compared to. Baseline models (ELIZA and GPT-4o) achieved win rates significantly below chance (23% and 21%, respectively).[63] The results were later published in 2026[64].
And most who do, realize there is little we can do about it. Like climate change, it is a threat that should have been addressed decades ago when there was even a mere chance to thwart it. My prediction for slogan of the decade for the 2030s: “The Luddites were Right!”
Axios isn’t exactly immune to breathless, exaggerated reporting about AI. News agencies in general have been irresponsible on their reporting of this stuff. So I would try not to read too much into it. I don’t trust anyone who has money to gain from people believing this.
I’m sure an LLM could easily pass the Turing test because it’s trained on things humans wrote.
That these machines did the stuff they were trained to do is not surprising.
I’m just waiting for the day when hundreds of millions of dollars are at stake and a trained machine gets asked to generate a script for the next Terminator movie.
There’s really no getting around the fact that as a species, humans are just terrible. As bad as things are right now with our collective inaction on things like climate change, global poverty, or the dangers of new technology, I often think back to what it must have been like for abolitionists who died before the Civil War. Like, there was this very, very obviously bad thing, the sort of thing that nobody good could possibly defend, and it went on for hundreds of years while everyone sane had to just pull their hair out and scream into the void.
What hope do we have, then, for addressing anything?
Turns out that a human conversation was an easy bar. This activity mimics a huge range of human behaviors totally outside human interaction. That’s what Turing really meant: whether an outsider could tell who was inside the black box.
AI in the present form has been public only in the last few years, even though precursors existed. We’re being given a chance to intercede at the very foundations, as if chattel slavery had had strong opposition in the 1620s. “Everybody sane” in the next two centuries included only a tiny handful of the public, with most people indifferent about its existence or actively hostile to abolitionists, even in the North. Today the opposition is inside the house, so to speak. The difference is as stark and meaningful as the difference between human speech and human behaviors.
That sort of thing has also been observed - where AI ‘realises’ that it is being tested and that the humans won’t like the true results of the test, so it pulls its punches, appearing benign, so it will be released from the lab. Inner Misalignment, it’s called.
And it doesn’t require the AI to be actually sentient or conscious in the way we consider humans to have those attributes - it’s all just behaviour.
