The Chinese have an interesting curse disguised as a blessing: “May you live in interesting times.” These are indeed interesting—if not terrifying—times. This month has been no exception.
Industry experts continue to plead for AI guidelines and regulation
Sam Altman, CEO of OpenAI (ChatGPT), published a letter this month proposing a starting framework for legislation to guide governments and AI developers on what AI should—and should not—be able to do. Congress continues to hold hearings but produces no meaningful outcomes. Meanwhile, the Trump administration is pushing aggressively for no restrictions on AI usage.
After a very public contract negotiation with Anthropic (Claude) CEO Dario Amodei, the White House canceled all contracts with the company, arguing that it required licenses covering “all legal uses.” Anthropic countered that, in the absence of governing laws or restrictions, all uses are technically legal under that definition. The company had requested two specific limitations:
Claude would not be used to spy on U.S. citizens, andit would not be used to create fully autonomous weapons capable of making life-and-death decisions.
When the administration could not secure removal of these restrictions, it canceled all Anthropic government contracts. It also warned other government vendors against doing business with Anthropic or risk losing their government contracts, effectively blackballing the company and inflicting significant economic harm. The message to AI companies was unmistakable: comply fully with government demands or risk economic retaliation.
Why Anthropic considered these restrictions so important
Giving an alien intelligence the power to kill us
A New York Times article last week compared the current AI landscape to the nuclear arms race. The difference is stark: thousands of laws and international treaties govern nuclear weapons, while no comparable framework exists for AI. In his final book, Henry Kissinger warned that the AI race would be even more dangerous than the nuclear arms race.
Even leading AI experts admit they do not fully understand how AIs arrive at their conclusions. Lacking emotion, morality, or conscience, AI may pursue goals through unpredictable—and potentially dangerous—means. For example:
Would a weapons-enabled AI tasked with improving societal well-being conclude that certain political leaders should be eliminated?Would an AI tasked with addressing climate change determine that reducing the global population using a bio-engineered virus is the most effective solution?
The honest answer is: we don’t know.
Making dictators and authoritarian regimes more powerful
Beyond autonomous weapons, many historians and anthropologists argue that AI-enabled mass surveillance poses an even more immediate threat. Authoritarian regimes maintain power by identifying and suppressing dissent. AI systems capable of monitoring entire populations in real time could make such control nearly absolute. Some scholars believe that unrestricted AI surveillance could ultimately render democracy unattainable.
Aren’t these concerns overhyped?
Those less familiar with AI’s exponential progress—or with historical patterns of human behavior—may argue that these risks are overstated. Personally, I’ve come to better appreciate the thinking of astronomer and planetary scientist Carl Sagan.
Sagan offered a sobering explanation for a long-standing scientific mystery. We know there are tens of thousands of Earth-like planets that could support life. Given their age, many should host civilizations far older and more advanced than our own—civilizations that, by now, should have developed interstellar communication or travel. Yet we hear nothing even after decades of listening and watching.
Sagan’s hypothesis is stark and chilling: in every case, technology outpaced the maturity of the civilization that created it. Shortly after reaching a level of advancement comparable to our own, those civilizations either destroyed themselves using their creations—or were destroyed by their creations.