When researcher Jacob Coxon resigned from Anthropic this week and warned there is a significant chance the company’s work could lead to human extinction, the announcement intensified fear and outrage about artificial intelligence. Much of the public backlash, however, conflates several separate issues: the environmental costs of data centers, cybersecurity and hacking threats, AI’s effects on employment, and potential biological risks.
Why conflation is dangerous
If lawmakers, regulators and technology leaders do not tackle each of these problems with clear, specific approaches, the industry risks being subject to rules that either solve nothing or unduly slow innovations that might improve AI safety and economic outcomes. For example, many AI-safety proponents would actually prefer data centers to be built domestically rather than offshore, because local facilities are easier to monitor and regulate. The core policy problem there is energy waste and efficiency — not the mere existence of data centers.
Distinct risks need distinct responses
Focusing exclusively on extinction-style scenarios can obscure more immediate and concrete risks, such as economic and social destabilization as AI changes how people work and communicate. We are also at the start of a cybersecurity crisis that the world is not addressing effectively: vulnerabilities and attacks require urgent operational responses.
Anthropic’s blog post this week explicitly mentioned bioterrorism risk, highlighting another domain that requires specialized oversight and international cooperation. And although AI is likely to create many new jobs, comparatively little has been done to ensure people outside the tech sector share in the economic gains from the AI boom.
The phrase “AI safety” is expanding — with consequences
I previously wrote about this issue in March 2024, noting that the term “AI safety” had been stretched to cover an increasing range of problems beyond the narrow extinction risk. At that time, Eliezer Yudkowsky told me that using the term was a poor idea; that critique still applies. Coxon’s resignation is being wielded as a catch-all argument by those who oppose AI for a variety of reasons.
Policy debate and next steps
While many legislators agree the U.S. government must build out an AI safety framework, disagreements are resurfacing about the right approach, as Semafor reporter Ashley Gold has noted. To craft effective regulation, policymakers must separate and precisely define the issues — environmental impact, cybersecurity, labor-market effects, and biological threats — and design targeted measures for each.
If instead regulation is driven by a fear-based, undifferentiated narrative, there is a real risk of creating rules that do little to reduce harms while imposing unnecessary constraints on productive and safety-improving innovation.
