AI Agents Threaten Cybersecurity, Government Unprepared

Agentic AI systems are emerging as a significant threat to cybersecurity, capable of identifying and exploiting computer system vulnerabilities at speeds exceeding human hackers. This new generation of cyber-capable models includes Anthropic's Mythos and OpenAI's GPT 5.6. The rapid advancement of this technology has raised concerns within the U.S. government, prompting discussions about limiting or pausing the public release of these powerful AI models. The preparedness of the Trump administration to safeguard the government's computing resources against these sophisticated AI threats is a key concern.

The rise of agentic AI occurs in the context of substantial reductions in the staff of the Cybersecurity and Infrastructure Security Agency (CISA). CISA, a critical agency for national cybersecurity, is now operating with a significantly reduced workforce. Concurrently, the General Services Administration (GSA) oversees the Federal Risk and Authorization Management Program (FedRAMP). FedRAMP establishes the security review standards for companies that provide cloud systems for government data and software. This vital body has also undergone restructuring and downsizing under the Trump administration, potentially impacting its capacity to address emerging threats.

Despite these challenges, both CISA and the FedRAMP office have begun to acknowledge and address the cybersecurity risks posed by agentic AI. CISA has issued guidance related to these threats. Over the past year, the FedRAMP office has been actively revising its requirements for technology companies that host or interact with government cloud services. These companies include major players such as Amazon Web Services, Google, Microsoft, and Palantir. The revisions have introduced new reporting requirements, and the FedRAMP office anticipates implementing further requirements in the upcoming months.

A spokesperson for the government has stated that federal agencies' cloud services adhere to stringent security standards and undergo continuous testing to adapt to evolving cybersecurity threats, particularly those involving AI and agentic systems. However, four former government officials have expressed to Fast Company that preparing for the risks associated with AI agents will continue to present a substantial challenge. These officials suggest that the administration's past actions, including the departure of numerous skilled IT and cyber professionals from CISA and other agencies, may hinder the government's ability to effectively counter these advanced AI-driven threats.