Speed and security are not the only concerns for AI: more defenders need access to frontier AI capabilities
Among those currently discussing the risks of AI, two views are dominant: the first, held by Anthropic’s CEO Dario Amodei and others, is that AI brings risks that warrant government regulation—risks of losing control of AI systems, of AI misuse for cyberattacks and bioterrorism, and of economic disruption. The second, held by free-market advocates and some skeptics, is that large AI companies (such as Anthropic and OpenAI) are exploiting their frontier status and U.S. Government connections to lock out competitors and create a monopoly for themselves.
AI luminaries, Jensen Huang and Mark Zuckerberg occupy a middle ground, believing government intervention could create new problems, slow innovation and ultimately increase risk.
Amodei’s recent essay, “We Must Pace the Frontier,” is a call to slow AI model development that is focused on cyber risks. Sam Altman, CEO of OpenAI, has pledged to adopt some of Amodei’s suggestions, and Elon Musk said, “Dario is right.”
I fully support many of Amodei’s proposed measures:
- Limiting the sale of powerful AI chips and semiconductor manufacturing to authoritarian regimes.
- Preventing companies in authoritarian countries from copying and recreating advanced AI models.
- Strengthening security at AI companies to prevent model weight theft.
However, efforts to slow AI model development do not seem realistic to me. As Amodei recognizes, meaningful constraints would require verifiable international agreements. Moreover, it is not clear that limiting AI progress would improve security given that defenders and the organizations they protect benefit from those same advances. A better path is greater collaboration and partnership between large language model (LLM) providers and security defenders to ensure defenders are equipped with frontier AI model capabilities.
This matters to every company, not just the frontier labs: attackers are already using AI to find and exploit vulnerabilities at machine speed. If frontier access is concentrated among a handful of firms, the benefits of advanced AI defense remain unevenly distributed across the economy. A more secure AI ecosystem requires that defenders have broad access to advanced AI capabilities. Broad access strengthens resilience across the economy and reduces the risk that weaknesses in one organization can have outsized consequences for others.
2026 Email Threats Report
Learn how AI and phishing-as-a-service are reshaping the email threat landscape and how to stay protected
Subscribe to the Barracuda Blog.
Sign up to receive threat spotlights, industry commentary, and more.
The Managed XDR Global Threat Report
Key findings about the tactics attackers use to target organizations and the security weak spots they try to exploit