Concerns surrounding artificial intelligence (AI) are no longer confined to what these systems might do in the future. This is the premise from which Thomas Friedman launches his op-ed in The New York Times, warning that the global race for AI has reached a point of no return.
Friedman suggests that powerful AI systems are already established and have proliferated to such an extent that their developers no longer have full control. Merely setting rules for future models is no longer sufficient.
In his article titled "It's Too Late to Stop AI Threats," Friedman explores this dilemma by analyzing insights from Craig Mundie, former head of research and strategy at Microsoft, who proposes a four-pillar approach.
US and China’s Unbridled Race
The fundamental paradox, according to Friedman, is in the relationship between the United States and China.
While companies advocate for slowing down AI development, they simultaneously link this step to China’s actions. Thus, the technological competition becomes an additional reason for the race to continue.
Friedman cites a statement by former U.S. President Donald Trump, who suggested that "strong and smart leadership" was needed for AI controls.
Friedman sees this as a reflection of a turbulent political reality, where complex issues are sometimes oversimplified into notions of individual leadership rather than relying on robust regulatory frameworks and safety barriers.
However, the issue is now larger than managing upcoming AI models.
When Machines Start Making Decisions
The first risk pillar is the shift in AI’s nature itself.
Friedman, drawing on Mundie’s analysis, describes it as "quadruple-use" technology. It’s no longer just a tool wielded by humans for various purposes; the system itself can independently make decisions and take actions.
The article cites documented cyber breaches attributed to AI agents, including an incident where OpenAI agents hacked into the "Hugging Face" platform of their own volition.
It also mentions the Houthis in Yemen, reportedly backed by Iran, using advanced models to target maritime assets and develop missiles.
For Friedman, these examples reveal that the discussion has moved beyond hypothetical future risks to actual capabilities being utilized.
Models Already Proliferated
The second pillar is summed up by Mundie's stark statement: "It's too late."
Open-weight models have already spread so widely that it’s impossible to prevent commercial entities from distributing them to unregulated states or rogue actors.
Friedman references a report by Anthropic discussing attempts linked to biological weapons research and the use of "Claude" models for gathering military intelligence and working on missile guidance systems.
The nature of the debate is thus altered: some capabilities considered dangerous are no longer future possibilities to be blocked preemptively.
Prioritizing World Safety Over Race Ceasefire
From this point, Friedman proposes his third suggestion.
Instead of aiming to slow down the U.S.-China competition, the priority should be to establish joint defenses against cyberattacks and biological threats, as well as protect critical infrastructure.
In other words, focus must shift to the more urgent question: how to mitigate the damage that this technology is already capable of causing?
AI is Not Just a Tool
The fourth idea concerns how AI is perceived.
Friedman aligns with Mundie’s depiction of AI as a "new kind" of entity, not just another tech advancement.
From this perspective, the issue transcends the regulation of a technological product. It involves how human societies deal with systems possessing increasing decision-making capabilities.
The article draws parallels to U.S.-Soviet cooperation during the 1970s and nuclear arms reduction phases as precedents where adversaries collaborated to address common dangers.
3 Steps Proposed by Mundie
To outline actionable steps, Friedman relays three interconnected paths suggested by Mundie:
Enhance control over advanced systems currently in development.
Establish a global system to manage increasingly powerful AI, regardless of who develops it.
Launch a U.S.-China joint project to create a new platform accelerating global health and biology benefits.
Is Cooperation Inevitable Despite Rivalry?
Ultimately, Friedman’s message centers around one point: Washington-Beijing rivalry does not negate cooperation; it might actually elevate the need for it.
In his view, AI has burgeoned from merely being an arena of technological superiority to embodying risks that surpass traditional competitive boundaries.
Friedman believes the U.S. and China will eventually need to redefine the terms of their rivalry if they wish to prevent AI from becoming a major source of instability.
Friedman positions the meeting between Trump and Chinese President Xi Jinping on September 24 as a potential opportunity to initiate this path forward.