AI companies have largely developed their chatbots with very little government regulation, all in an effort to beat China to artificial general intelligence. However, as these services exploit users’ mental health, enable new devastating cybersecurity threats, and arm the U.S. military with advanced capabilities, the Trump administration recently proposed federal regulation to keep the bots in check. Now, President Donald Trump wants to take government oversight a step further by invoking the power to review AI models before they’re released to the public through an executive order that was signed this week.
President Trump has mostly maintained a hands-off approach to AI regulation, bucking attempts at state-level bills to curb development in favor of a centralized federal mandate. There are clear pros and cons to Trump’s National AI Legislative Framework, but it provides a starting point for standardizing an industry where Trump has dragged his feet.
This is why the latest reports of added AI oversight, directly from the U.S. government, come as a surprise, given Trump’s previous stance. If signed, the executive order would mark a sea change within the Trump administration, signaling that AI needs direct government intervention to protect the public from potentially dangerous models.
The question is, why?
This move raises the question: How much AI regulation is too much regulation?
Trump’s decision came after Anthropic — the same company that landed on the military’s supply chain risk list — unveiled a new AI model that was purportedly too dangerous to release to the public. Labeled as Mythos under Project Glasswing, the new model excels at leveraging computer hacking and cybersecurity exploits. In other words, it’s really good at breaking the security measures of critical digital products and services, including operating systems and internet browsers.
If left in the wrong hands, Mythos could pose a huge risk to anything and everything connected to the internet — personal devices, school computers, government systems, banking platforms, and even critical infrastructure like power grids, traffic systems, and more.
Instead of allowing the public to access Mythos outright, Anthropic opted to provide the model strictly to Big Tech companies to help them find security holes in their products before a competing AI platform on the same level as Mythos reaches public status. The goal is to patch these bugs before they are exploitable by hackers using other AI platforms. So far, Mythos has poked holes in Apple’s highly secure MacOS platform and Mozilla’s privacy-focused Firefox browser. Unfortunately, while Mythos is good at finding problems, it’s bad at patching them, with recent reports noting that Mythos can further break software, even when trying to fix it.
Not to be outdone, OpenAI also claims to have a model — GPT-Rosalind — that’s too powerful for public release, this time in the sector of life sciences and molecular biology. Instead of launching Rosalind broadly, the company is offering it to researchers and scientists only.
So far, Anthropic and OpenAI have been socially responsible with their models by self-limiting access, but there’s no mandate to enforce these restrictions. President Trump’s executive order aims to eliminate any leeway and prevent truly dangerous AI models from leaking into the mainstream.
RELATED: The Trump phone is here — and so is the controversy. Is it any good?
BasSlabbers/Getty Images
This move raises the question: How much AI regulation is too much regulation, and what are the ramifications of government overreach on access to the most advanced technology known to mankind? Some view these bills and mandates as a danger to free speech. Others see it as a government power grab meant to control device, internet, and AI access. I’m somewhere in the middle — the government should prevent AI companies from outright harming the people while also keeping the people’s rights and freedoms intact.
Unfortunately, even if the Trump administration has the best intentions with its AI executive order, who’s to say that the next administration will be so benevolent? Direct government intervention over AI models gives the left the precedent they need to overtly regulate and even manipulate AI the next time they take power. Imagine a future where the left blocks AI models on the grounds of “misinformation” and “disinformation” for sharing facts that don’t align with their political views. It’s not like they didn’t try to wipe dissent from the internet before, and if given the chance, they’ll do it again.
Luckily, the left might not get that opportunity. President Trump’s AI executive order was put on hold the day it was meant to be signed, though the unsigned version was leaked online for your viewing pleasure. Still, even with the order paused at the eleventh hour, its albatross looms as a possibility for future AI regulation that could either save the people from certain chatbot destruction or steal away our rights to access “unapproved” versions of these models that don’t comply with the party in power.
Tech, Artificial intelligence, Trump administration, Executive order, Anthropic, Openai
