“I got calls from people inside of some of the AI labs and it felt like getting a call from Robert Oppenheimer before the atomic bomb.”
- Tristan Harris, episode 449 of Talk Easy with Sam Fragoso
The idea of an AI Cold War that’s been bubbling for years now has reached a fever pitch this fall as the United Nations Security Council met with AI leaders in New York City while President Trump hosted President Xi Jinping of China in Washington, D.C.
I’m supportive though skeptical of OpenAI’s “automated shutdown” and Anthropic’s “pacing the frontier” efforts (even if they only came about from highly publicized breaches and fleeing employees). But these organizations, along with Meta and Nvidia, and others, will continue to move the goal posts as it suits their business and political interests. Perhaps the UN will manage to learn from the original Cold War and successfully align nations for the greater good. But while global compacts are arranged and kill switches generated, we’re still in the midst of managing boundless technology that’s been integrated into everything from finance software to air conditioning units. We are all still interfacing with AI every day so we must also be the arbiters of its ethical use.
To a CTO, department head, or the lone director leading burgeoning AI efforts, that might seem like an overwhelming responsibility. But we have frameworks, expertise, and industry experience; your organization has a vision and core values and rules of engagement. Think about the world you want to build and the policies and boundaries you would like to exist and we can shape AI to fit those systems.
For some organizations that might mean limiting AI’s interaction with live data or regularly scheduled vetting of ATS algorithms or keeping it out of compensation decisions entirely. To others it might mean prioritizing vendors whose AI agents are independently validated and partners who aren’t passively adopting new models into their third-party systems without doing appropriate vetting.
Dario Amoedi’s proposal for independent organizations to provide verifiability, transparency, and a second opinion for Anthropic's AI models is one of the first use cases I imagined for a partnership like the one we’ve been offering with TrustVector since our very beginning back in 2021. It’s nice to see that the frontier AI leaders are catching up to what the responsible AI community has been saying and doing for quite some time. I can see how easy it is to get caught up in building AI systems that are cutting-edge, fast, and constantly improving. But what they really need to be is practical, sustainable, and defensible long-term.
To state the obvious: The tech is moving quickly; the regulation has been—and will continue to—develop slowly. Australia’s defense minister and deputy prime minister, Richard Marles said that the government was still trying to determine whether laws were breached when OpenAI’s agent erroneously hacked a government healthcare portal this summer.
In a recent video on the Trump-Xi summit, David E. Sanger said, “Being concerned isn’t a policy and it’s going to take a lot of long and laborious negotiations to come up with any kind of common ground and even then there’s a question about how enforceable it would prove to be.”
Instead of treading water in the open seas of the whims of AI oligarchs and asking for life rafts of forgiveness (or government intervention) later, let’s start setting expectations now. If you can’t rely on the pioneers of the technology to build a responsible tool, you have to be committed to using it responsibly.
Aleks Jevtic is CEO and founder of TrustVector. He’s a former PwC digital transformation executive with 20+ years of experience across healthcare, life sciences, enterprise technology, and AI risk.
