The Trump administration is deeply divided over whether U.S. intelligence agencies should gain expanded authority to evaluate artificial intelligence models, revealing fundamental disagreements about how to balance national security concerns with technological innovation, according to two people familiar with the matter who spoke on the condition of anonymity to discuss a proposal that is not yet public.
The internal conflict comes as the White House confronts mounting cybersecurity threats posed by advanced AI systems. National security officials are pressing for greater influence over AI regulation, while other parts of the administration appear to resist expanding intelligence agency power in this domain. David Sacks, former White House AI czar, remains active in ongoing discussions about how the administration should respond to AI advances.
The Security Testing Gap
U.S. officials attempted to conduct comprehensive security tests on AI systems developed by Microsoft, Google, and xAI to identify potential threats ranging from cyberattacks to military misuse. However, details about these stress-tests were subsequently deleted from a government website, and the reason for the removal remains unexplained. The deletion raises questions about transparency in how the government evaluates the security risks posed by increasingly powerful AI systems from major technology companies.
The removal of publicly available information about security testing protocols underscores the tension between transparency and security concerns that characterizes the current regulatory environment. Citizens and civil society organizations typically rely on publicly disclosed information to understand how government agencies oversee emerging technologies that could affect national security and public safety.
Dual-Use Technology and Criminal Exploitation
Google disclosed that criminal hackers successfully used artificial intelligence to identify a major software flaw, demonstrating the dual-use nature of AI technology. The company's revelation illustrates that the same tools and capabilities that can strengthen cybersecurity defenses can also be weaponized by malicious actors to compromise critical systems.
Google emphasized that the field remains in its early stages and requires sustained, coordinated effort to develop safer code and more robust security practices. The company's assessment suggests that without deliberate investment in security infrastructure and oversight, the risks posed by AI systems may outpace the protective measures currently in place.
Why This Matters:
The administration's internal divisions over intelligence agency oversight of AI models reflect a broader policy challenge: how democracies can effectively regulate transformative technologies while maintaining public accountability and transparency. When government security testing is conducted but details are deleted without explanation, the public loses visibility into how powerful institutions are managing risks that could affect millions of people. The fact that criminal actors are already exploiting AI capabilities to find software vulnerabilities demonstrates that regulatory frameworks must keep pace with technological advancement. The current debate will likely determine whether AI safety oversight becomes more centralized under intelligence agencies—raising questions about public input and democratic accountability—or remains fragmented across multiple agencies with unclear coordination. These decisions carry significant implications for cybersecurity, technological competitiveness, and the balance between security and transparency in a democratic system.