
Former U.S. National Security Advisor: "An average hacker can inflict severe damage, and we are not prepared"
Jake Sullivan, who served as national security advisor under President Biden, warned at the New Era Capital conference in Boston about the growing threat posed by the intersection of artificial intelligence and cyberattacks on critical infrastructure. He said even a six-month lead over China in the AI race is significant, particularly as the U.S. struggles to protect its power, water and communications networks. He also predicted that China would restrict access to its most advanced AI models by 2027.
“What is currently available on the dark web, enabling an entirely average hacker to inflict severe damage on critical global infrastructure, is staggering. And we are not prepared to defend ourselves.”
With this stark warning, former U.S. National Security Advisor Jake Sullivan opened his keynote address at the New Era Capital conference in Boston. Sullivan, who served as the 28th U.S. National Security Advisor under President Joe Biden from January 2021 to January 2025 and currently holds a professorship in statecraft and world order at Harvard University, presented a detailed analysis of the geopolitical balance of power, the artificial intelligence race with China, and cyber threats to critical infrastructure. Sullivan recently joined the advisory board of New Era Capital.
Sullivan began by explaining how the Biden administration's policies on technology and cybersecurity took shape. Just months after the administration took office in 2021, a ransomware attack crippled the Colonial Pipeline, causing severe fuel shortages along the U.S. East Coast. The incident, he said, exposed the government's limited legal authority to compel private companies operating critical infrastructure, including water systems, communications networks and airports, to meet basic cybersecurity standards. The administration's subsequent requests to Congress for such powers were rejected.
Toward the end of 2021, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei met with Sullivan separately and told him they were developing a transformative technology that would rely on enormous computing power. Although he initially viewed their claims with skepticism, the administration came to recognize during 2022 that computing capacity would be central to winning the technological race. In October of that year, roughly a month before ChatGPT was launched, the administration introduced its first major round of export controls on advanced AI chips and the equipment used to manufacture them, targeting China's access to the technology.
Sullivan also recalled a forecast Amodei shared with him in early 2023. Amodei predicted that within three years, AI models would be capable of identifying and exploiting vulnerabilities in large-scale computing systems. Sullivan said Anthropic's Mythos model, released in the spring of 2026, matched that projected timeline and those anticipated capabilities.
Responding to the argument that a lead of only six months over China was not particularly significant, Sullivan offered a different perspective.
“If China possessed the Mythos model six months ahead of the U.S., and were able to deploy it, and then, six months later, unveiled a new model that still kept them six months ahead of the U.S., would we be concerned? As National Security Advisor, would I be able to sleep at night without being deeply worried? The answer is: damn right, I would. That is why six months is critical, especially regarding the balance between offense and defense in cyberspace.”
Turning to the current balance of technological power, Sullivan addressed recent reports concerning Nvidia. Beyond the allegations of espionage and efforts to lure away talent, he said, the central issue was “a story of industrial-scale circumvention of chip controls, which granted China access to a massive amount of American computing power and narrowed the gap to some extent.”
Nevertheless, Sullivan argued that Nvidia alone currently produces approximately 25 times more computing power than China's Huawei, a comparison that does not include the output of companies such as AMD, Google and Amazon. He estimated that the gap between the two countries would remain substantial through 2030, barring an unexpected breakthrough by Beijing in semiconductor manufacturing.
Sullivan said China's efforts to circumvent U.S. restrictions rely on remote access to computing resources and smuggling. He predicted that Congress would move to close these loopholes by the end of the year. He also pointed to a new Department of Commerce directive restricting remote access to advanced chips, which he said had been drafted with direct involvement from the Treasury Department and the National Security Council.
China's broader strategy, he argued, is to compete through lower prices and open-weight AI models, distributing “good enough” systems that could eventually become the global standard. Sullivan warned that the combination of closed, frontier-level models and widely available open models creates a particular security risk: the safety mechanisms built into open models can be stripped away, potentially turning them into powerful cyberattack tools accessible to a much wider range of users.
Sullivan identified three issues that are increasingly preoccupying policymakers: the evolution of China's open models, the balance between cyber defense and offense, and the question of deterrence.
On the first, he said China's internal debate has recently shifted away from an emphasis on openness and toward greater security and control. He predicted that by 2027, Beijing would likely restrict distribution of its most advanced open models, either limiting access to domestic users or introducing a tiered system of access. Establishing limits on the capabilities of such models, he added, should be a key U.S. diplomatic objective in its dealings with China.
On cyber defense, Sullivan argued that initiatives designed to protect systems, such as Anthropic's Glasswing and OpenAI's Patch the Planet, may help large corporations but do not address the vulnerabilities of local infrastructure operators, including power grids, water systems and airports.
He pointed to the Salt Typhoon operation, in which cyber actors linked to the Chinese government penetrated the networks of all major U.S. telecommunications providers. Removing them required months of intensive work, he said, yet the Federal Communications Commission did not impose binding cybersecurity requirements on the companies, citing the high costs involved and the absence of a clear legal mandate. Sullivan also referred to a proposal by former U.S. Deputy National Security Advisor for Cyber and Emerging Technology Anne Neuberger to disconnect critical industrial control systems from the internet, reducing their exposure to cyberattacks.
Sullivan acknowledged that the United States still lacks a formal cyber-deterrence doctrine comparable to those governing nuclear or conventional warfare. He cited China's Volt Typhoon campaign, in which actors linked to the Chinese military embedded malware in U.S. critical infrastructure in preparation for a potential conflict over Taiwan.
In response, Washington made clear to Beijing that activating malware in a way that caused physical destruction to power grids, pipelines or railways would be treated as a physical military attack and would warrant a corresponding response. However, Sullivan noted that the U.S. remains vulnerable to more routine cyberattacks, including those carried out by Russia and Iran. Its highly interconnected and decentralized infrastructure makes it difficult to defend, and deterring such attacks remains an unresolved challenge.
In his concluding remarks, Sullivan turned to the role of cyber warfare in modern conflicts. Before Russia's invasion of Ukraine, many forecasts predicted that Moscow would use a preemptive cyberattack to neutralize Ukrainian forces. That did not happen. Instead, drones, powered by software and artificial intelligence, proved to be among the technologies that most dramatically changed the battlefield.
Sullivan argued that conflicts over the coming decade would be defined less by large-scale, integrated ground maneuvers and more by a new triad: special operations forces, space and cyber capabilities. Under this model, the physical operations of special forces would be coordinated with cyber effects, while space-based assets would provide intelligence and advanced sensing capabilities.














