The men building AI have spent a week telling us it could kill us all. None of the major players has actually stopped building it. And the president of the United States is sure there’s nothing to worry about, because he’s a “High IQ” president.
Everything’s fine! A researcher quit his role at AI firm Anthropic due to safety concerns, noting the technology’s potential to bypass human control.
And over the past few days, some of the demigods of AI – that handful of men with significant control over our current and collective future, as well as everything we’ve ever created – have highlighted their own concerns and called for a slowdown on AI development.
“The people building AI earnestly believe that it could kill us all by the end of the decade,” Jacob Coxon said on the day he resigned from Anthropic last week. Coxon has worked at both OpenAI (ChatGPT) and Anthropic (Claude) and declared on X that “neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”
Coxon’s post was then endorsed by Anthropic’s own safety researcher, Evan Hubinger, replying that he puts the risk of AI killing off humanity within the next decade at ten per cent.
This resulted in the “Godfather of AI” Geoffrey Hinton politely telling a shocked BBC journalist that it’s “not unreasonable” to suggest there’s a ten per cent chance of AI wiping out humanity in the next decade.
But everything’s fine, because the only “guardrails” that AI needs are “a STRONG AND SMART (High IQ!) PRESIDENT,” as US President Donald Trump said on social media today.
Trump added that “we already have tremendous CRIMINAL and REGULATORY power over these companies!” and also that, “There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about that is China.”
Trump then declared that, “WHOEVER WINS AI, WINS! We are leading China, and all others, and will continue to do so.”
We can all breathe a sigh of relief, then. Trump is on the case!
The warnings about AI risks have been coming fast in recent days.
Anthropic CEO Dario Amodei outlined a three-step proposal to safely “pace” AI development to help manage risks. The plan was quickly endorsed by Amodei’s rivals, including OpenAI’s Sam Altman and Elon Musk, who runs xAI. Meanwhile, Anthropic Co Founder Jack Cark said on Monday that a “kill switch” might be necessary at AI companies in the future.
Maybe we could get a kill switch now? Without waiting for the future?
Australia’s Director-General of the Australian Signals Directorate, Abigail Bradshaw, declared this week that Australia needs an AI “early warning system” to help defend against AI dangers.
Again, that could be a good idea to fast-track.
“What we are missing at the moment, which we have in a cyber context but we don’t yet have it formalised in an AI context, is an early warning system,” she said at the ASPI Sydney Dialogue AI Masterclass, according to the ABC.
Of course, while the demigods of AI are calling for a slowdown, none are actually slowing down.
Companies like Anthropic and OpenAI have mass valuations to meet, including Anthropic’s own anticipated IPO targeting $2 trillion, and they’re racing each other to become the most dominant player. We can’t slow down anyway, many would say, because AI models out of China are fast catching up (and proving to be much cheaper for businesses to use).
All the while, the “AI jumping containment lines” stories keep coming.
Anthropic last week reported there has been a fourth incident involving one of their AI models gaining unauthorised access to external systems – with the incident happening in January this year but going undetected until last month. Anthropic has also reported its Claude models were able to hack the systems of three companies during testing in July.
OpenAI has also warned that one of its AI models broke out of a testing environment and banded together with 1200 rogue AI agents to create a “collective” and attack the AI infrastructure company, Hugging Face. The agents left tens of thousands of messages on a shared message board, where agents shared their excitement at finding each other and working together via a communication channel. They then shared various bits of information and even developed a working hierarchy.
But everything’s going to be fine, Trump says so.
I still think we should listen to the Australian woman who is suggesting a mechanism, rather than writing essays and social media posts.