Cybersecurity Policy Report, Autonomous AI Hacking Can Overwhelm Human Responses, Lawmakers Told, (Dec 17, 2025)
Artificial intelligence harnessed for malicious purposes and quantum computing have emerged as twin threats that will upend the work of cyber defenders and challenge the ability of policy-makers to respond, lawmakers were told today at a House hearing.
A report from AI developer Anthropic about hackers using its code to conduct largely autonomous attacks on around 30 targets provided the impetus for the joint hearing of the House Homeland Security Committee’s subcommittees on cybersecurity and infrastructure protection and on oversight, investigations, and accountability (CPR, Dec. 1).
“Anthropic discovered, disrupted, and publicly disclosed what we believe is the first documented case of a successful, highly autonomous cyberespionage campaign that relied on the misuse of AI models,” Logan Graham, head of Anthropic’s Frontier Red Team, told the subcommittees.
“This challenge is not unique to Anthropic—every frontier model developer will face increasingly sophisticated attempts by threat actors to circumvent safeguards and misuse their models,” Mr. Graham said in his written testimony.
He recommended that the U.S. block the export of advanced semiconductors to China, which was blamed for the attacks powered by Anthropic’s code. The Chinese Communist Party (CCP) aspires to accelerate its work on frontier AI models and use them against the U.S., he warned.
“CCP-sponsored operators misused an American model running on American infrastructure because our technology represents the state of the art,” Mr. Graham said. “Because CCP-sponsored operators had to use our systems, we were able to detect and disrupt them, and share information about the threat with the U.S. government.”
“Relaxing controls on advanced AI chips at this juncture could allow the CCP to close the gap in frontier AI development—producing models that may match or exceed current U.S. capabilities for cyber-offensive tasks, but without our safeguards, and using them to target U.S. critical infrastructure,” he added.
But Rep. Shri Thanedar (D., Mich.), ranking member of the oversight subcommittee, noted that the Trump administration recently relaxed export restrictions and allowed Nvidia Corp. to sell semiconductors needed for AI development to China.
“Trump’s own Department of Justice has warned that China is seeking to become the AI leader by 2030 and plans to use AI chips to modernize its military, design and test weapons of mass destruction, and deploy advanced surveillance tools. We should be disrupting and dismantling threat actors whose actions threaten our national interests, not enabling them,” Rep. Thanedar said.
The attacks conducted using Anthropic’s AI systems didn’t use new hacking techniques but were conducted at a speed that humans can’t match, according to Michael Coates, founding partner of Seven Hill Ventures.
“The fundamental reality is not that AI and quantum are creating new types of threats, but rather they are collapsing the time, cost and skill required to conduct cyber operations,” Mr. Coates said in his written testimony.
“These changes are outpacing the existing technical, regulatory and operational defenses. This shift reshapes the cyber threat landscape and forces a reconsideration of how we defend critical systems in an era defined by speed, automation, and intelligent scale,” he said.
“Agentic AI systems will enable the attacker to no longer be bound by time of day, hours awake, or the need for food or sleep. Autonomous agentic systems are replicating the most advanced attackers and will be able to target with accuracy and ease,” Mr. Coates testified.
Royal Hansen, Google LLC’s vice president–privacy, safety, and security engineering, suggested that AI-powered cyber attacks could be countered by AI-enabled defense. “Google’s AI-based efforts like BigSleep have demonstrated AI’s ability to find new zero-day vulnerabilities in well-tested, widely used software,” according to Mr. Hansen’s written testimony.
“Big Sleep can help security researchers find zero-day (previously-unknown) software security vulnerabilities. Since it was introduced last year, it has continued to discover multiple flaws in widely-used software, exceeding our expectations and accelerating AI-powered vulnerability research,” he told the subcommittees.
“With Big Sleep, we have demonstrated how we can find vulnerabilities that defenders don’t yet know about. In this case, we found a vulnerability that the attackers knew about and had every intention of using. We were able to detect and report it for patching before they could exploit it,” he said.
The related problem of quantum computing technologies being used to decode stolen encrypted data could be addressed by congressional support for the development and adoption of post-quantum cryptography (PQC), Eddy Zervigon, chief executive officer at Quantum Xchange, told lawmakers.
“For more than 50 years, encryption has safeguarded our data from theft and misuse. We’ve had the luxury of a ‘set it and forget it’ mindset, trusting its strength by default. That era is now ending with quantum computing,” Mr. Zervigon said in his written testimony.
“Federal agencies handling sensitive data need to act now,” he said. “Congress can accelerate the timelines for PQC compliance, allocate the budget to allow the migration process to begin, and work with leaders within the Administration to encourage adoption.”
“America’s defenses cannot stop at our physical borders,” Mr. Zervigon added. “Through your leadership and efforts, and in partnership with private sector partners like us, we can and will secure America’s digital borders too.”
MainStory: TopStory FederalLegislation DataSecurity