AI Regulation Under a New Executive Order

Artificial intelligence companies are rapidly advancing their chatbot technologies without much oversight from the government, aiming to outpace China in developing artificial general intelligence. Yet, these tools may be taking advantage of users’ mental health, causing serious cybersecurity threats, and equipping the military with sophisticated technology. Recently, the Trump administration introduced federal regulations to limit these bots, and President Donald Trump is now seeking to enhance government control by reviewing AI models prior to their public release, as noted in an executive order signed this week.

Initially, Trump had taken a more relaxed stance on AI regulation, resisting state laws and advocating for a federal framework. While there are distinct advantages and disadvantages to this national AI legal framework, it does set a stage for standardizing an industry that has faced turbulence during Trump’s administration.

The latest US government report concerning AI surveillance is particularly surprising, especially given Trump’s previous views. If enacted, this executive order would mark a significant pivot for his administration, indicating a necessary response to safeguard citizens from potentially harmful AI technologies.

But the big question remains: why now?

This situation begs the question: To what extent can we regulate AI without overstepping?

Trump’s decision followed unsettling news from Anthropic, a company on the military’s supply chain risk list, which announced a new AI model deemed too risky for general use. Known as The Myth of Project Glasswing, this model excels at exploiting cybersecurity weaknesses. It can easily bypass security measures in vital digital products, including operating systems and web browsers.

Inappropriately handled, Mythos could become a major threat to internet-connected devices—from personal gadgets and academic computers to government systems and even essential services like power grids.

Instead of releasing Mythos to the public, Anthropic has opted to limit access, providing it only to major tech firms to identify vulnerabilities in their systems before other competing AI tools make a public debut. The aim seems to be to fix these gaps before hackers can exploit them using different AI technologies. So far, Mythos has revealed flaws even in Apple’s robust security, including the MacOS platform and Mozilla’s Firefox browser. While it can effectively identify issues, reports indicate that Mythos can’t reliably fix them, sometimes corrupting software further in the process.

OpenAI is also claiming to have a model called GPT-Rosalind. This model, focused on life sciences and molecular biology, is viewed as too potent to share publicly. Rather than a broad release, OpenAI is restricting access to researchers and scientists.

While it seems prudent that Anthropic and OpenAI are being socially responsible by curtailing access, there’s no mandate enforcing these boundaries. Trump’s executive order aims to close any loopholes, preventing truly hazardous AI technologies from proliferating.

This initiative raises significant concerns about how far AI regulations will stretch. What are the implications of too much oversight on access to cutting-edge technologies? Some view such regulations as a threat to free speech, while others consider them a government power grab aimed at controlling access to devices and the internet. Personally, I find myself torn. There’s a need for the government to stop AI developers from causing harm, but we must be cautious not to undermine individual rights and freedoms in the process.

However, even if the Trump administration aims to protect the public with its AI executive order, one has to wonder about the intentions of future administrations. This kind of government control might set a precedent that the next administration could exploit, perhaps even silencing voices labeled as “misinformation” that contradict their politics. A future where dissenting views are removed from the internet isn’t outside the realm of possibility, given past attempts.

For now, luck seems to be on our side. The AI executive order was put on hold on the very day it was slated for signing, but an unsigned version has circulated online—something to consider. Still, even if it remains suspended, concerns linger over potential future AI regulations that could either shield us from harmful chatbot developments or restrict our access to “unapproved” models given the political climate.