Three senior US senators have formally demanded investigations into reported artificial intelligence failures in American military and intelligence operations, warning that unreliable data and AI-generated falsehoods risk causing lethal operational mistakes. The demand, directed at Defence Secretary Pete Hegseth and Director of National Intelligence Jay Clayton, calls for inspectors general to be granted unrestricted access to examine the reported incidents.

What the Senators Alleged

Senator Mark Warner, the senior Democrat on the Senate Intelligence Committee, Senator Jack Reed, ranking member of the Senate Armed Services Committee, and Senator Chris Coons, who leads Democrats on the Senate Appropriations defence subcommittee, co-signed the letter. They cited two specific public reports that, they argued, illustrate a dangerous pattern in the deployment of AI within military systems.

The first concerned an AI targeting platform that allegedly relied on outdated geospatial data during a US kinetic strike in Minab, Iran. The second involved an aborted military interdiction operation, reportedly halted after AI-generated false information — described as a potential 'hallucination' — was included in intelligence disseminated by US Special Operations Command Pacific. The senators noted that the reported failures have not been independently established in material released by them, and they provided no further operational details.

Key Concerns Raised

In their letter, the senators warned that 'AI targeting platforms have repeatedly generated spurious outputs based on outdated input data or outright hallucination,' a pattern they said risks serious operational error. They argued that such failures could undermine confidence 'in the accuracy and reliability of US intelligence and intelligence-informed kinetic actions.'

They also flagged what they described as Pentagon pressure on a leading American AI company that had reportedly resisted allowing its technology to be used in fully autonomous lethal operations. The company was not identified. Critics argue this backdrop makes transparent oversight all the more urgent.

Governance vs Speed: The Core Dispute

The senators criticised the current administration's revocation of National Security Memorandum 25 and a Department of Defense strategy which, they contend, treats governance requirements as obstacles to faster AI adoption. They described the approach as akin to a 'Move Fast and Break Things' mindset — suited, they wrote, to social media platforms but dangerously misaligned with military operations. This criticism reflects a broader tension in Washington over how rapidly AI should be integrated into national security infrastructure without adequate guardrails.

Notably, this comes amid a wider global debate over autonomous weapons systems. Several allied governments have pushed for international norms on AI in warfare, while the US military has continued to expand AI use in surveillance, pattern recognition, and operational decision support.

Implications for India and Allied Partners

The controversy carries direct relevance for India and other American defence partners, as militaries deepen cooperation in intelligence-sharing, surveillance, and emerging technologies. If AI-generated assessments cannot be independently verified before influencing military action, confidence in shared intelligence systems could erode. For India — which has progressively integrated defence and intelligence frameworks with the US under various bilateral arrangements — the reliability of American AI systems is not merely an academic concern.

What Comes Next

The senators have called for immediate investigations by relevant inspectors general, broader public disclosure of findings, and a review of whether other significant AI-related targeting failures occurred without public disclosure. Whether Hegseth or Clayton will comply — and to what extent — remains to be seen. The outcome could shape how the US military balances innovation speed against accountability standards across its AI programmes.