WASHINGTON, November 3, 2026 — OpenAI has appointed former US Secretary of the Army Dan Driscoll to a newly created senior counselor role focused on national security modernization, adding a former Pentagon leader with military and diplomatic experience to its work with government and defence partners.
Driscoll will advise OpenAI’s leadership and work with external partners on ways advanced artificial intelligence could support national security and cybersecurity, the company said in an announcement reported by Reuters late Monday. He is also expected to engage government officials, industry leaders and academic institutions on what OpenAI describes as “democratic AI” and help internal teams understand the operational requirements of national security organisations.
The announcement did not identify a specific new contract, deployment or classified programme tied to Driscoll’s appointment. It also did not detail the role’s reporting structure or the limits of his authority. The immediate change is a senior advisory position intended to connect the company’s AI work with the practical needs of national security users.
Driscoll said he was joining OpenAI to help the United States and its allies use the technology responsibly for people who keep them safe. The statement, quoted by Reuters, framed the role around cooperation with public institutions rather than a single product announcement. OpenAI said Driscoll would help its teams understand how national security organisations operate and what they need from advanced AI systems.
That remit could cover a broad set of issues: how agencies evaluate AI tools, what safeguards are needed when models process sensitive information, how systems can be integrated into existing workflows and how public-sector customers assess reliability. Those are potential areas of work, not a confirmed list of programmes. The company has not said whether Driscoll will oversee specific government accounts or take part in procurement decisions.
The phrase “democratic AI” is OpenAI’s description of its approach, not a formal legal standard. In practice, government use of AI raises questions about who controls the systems, what data they can access, how outputs are checked and whether people can challenge consequential decisions. Those questions matter whether a model is used for cyber defence, logistics, translation, intelligence analysis or administrative work.
National security customers also operate under constraints that differ from those of ordinary commercial users. Some systems must function in restricted networks, handle classified material or continue working when communications are unreliable. Procurement can require lengthy security assessments, clear audit trails and evidence that a supplier can maintain a system over time. A senior adviser with experience inside the Army may help a technology company understand those demands, although the appointment alone does not demonstrate that any particular capability has been approved for operational use.
Driscoll served as the 26th Secretary of the Army after being sworn in during February 2025. The US Army’s own account of his tenure describes a focus on readiness, procurement and modernising the force. Before entering senior civilian leadership, he served as a cavalry scout in Iraq. That experience gives him familiarity with the practical conditions under which military personnel use equipment, communications and information systems.
His time at the Pentagon also included an unusual diplomatic assignment. Reuters reported that President Donald Trump selected Driscoll to take part in talks with Moscow and Kyiv aimed at ending the war in Ukraine, a task not normally associated with the Army secretary’s portfolio. The assignment placed him in discussions where military requirements, political objectives and the limits of a negotiated settlement intersected.
Driscoll left the Army secretary position earlier in 2026 after months of tension with Defence Secretary Pete Hegseth, according to Reuters. The available reporting does not establish that his departure was related to the OpenAI appointment, and the company has not described it that way. His move from government to a technology company does, however, illustrate how people with defence experience are being recruited as AI firms deepen their engagement with public-sector customers.
For a company selling advanced models, that experience can be useful in translating a general capability into a service that meets an agency’s operational requirements. It can also help identify where a model’s limitations would make it unsuitable. Military and intelligence organisations need to know not only what an AI system can do in a demonstration, but how it behaves when information is incomplete, instructions conflict or a failure could affect a real-world decision.
The appointment does not make Driscoll a government representative at OpenAI, nor does it mean that the US military has endorsed the company’s products. He is joining a private company, and any government deployment would still depend on the relevant contracting, security, legal and operational review processes.
Also read: OpenAI Scraps GPT-6.1 Astra Release After Safety Tests
AI tools can help people sort large volumes of documents, translate material, identify patterns in network logs and draft software or reports. In cyber defence, they may help analysts prioritise alerts, investigate suspicious activity and understand how a vulnerability could affect a system. These uses can reduce repetitive work, but they do not remove the need for skilled staff to verify results and decide what action to take.
Security organisations also have to account for the same technology being used by attackers. Models can assist with reconnaissance, social engineering and code generation, while systems connected to sensitive networks may become targets themselves. A deployment that is helpful in a controlled environment can create risk if it has broad access to data, can execute actions without approval or cannot be reliably monitored.
That is one reason the boundary between a model’s ability and its authority is important. A system that summarises intelligence reports is different from one that can send instructions to operational networks or trigger a response. Human approval, logging, access restrictions and testing need to reflect the consequences of the task. The more consequential the action, the less appropriate it is to treat a model’s output as self-validating.
AI vendors also face a challenge in explaining how their systems were tested and what happens when they fail. Public-sector buyers may ask for information about model behaviour, data handling, retention, update procedures and incident response. Some of that material can itself be sensitive, but agencies still need enough evidence to assess the technology before relying on it.
OpenAI’s national security work is part of a wider competition among AI developers to serve government and defence customers. The commercial opportunity can be substantial, but the contracts often involve longer sales cycles and more demanding security requirements than consumer products. Companies must balance the desire to make advanced capabilities available with restrictions on how those capabilities can be used.
Also read: US Appeals Court Upholds Pentagon Blacklist of Anthropic
One unresolved question is how OpenAI will define the boundaries of its national security work. The announcement does not specify which agencies Driscoll will engage, whether his remit includes classified deployments or how the company will handle requests that raise legal or ethical concerns. Those details matter because “national security” covers a wide range of activities, from protecting government networks to intelligence analysis and military operations.
Another question is how the company will demonstrate that safeguards work in practice. Policies can set limits on the use of AI, but organisations also need technical controls that prevent unauthorised access, record consequential actions and allow operators to stop a system when its behaviour becomes unreliable. Independent testing and clear incident reporting can help distinguish a robust deployment from a promising demonstration.
Governments have their own responsibilities. Agencies must decide which tasks are appropriate for automation, who remains accountable for decisions and how procurement rules apply when models change frequently. They also need to assess the risks of depending on a small number of suppliers for foundational technology. A private company’s safety commitments do not replace the legal obligations of a government agency using its products.
There are wider questions about public trust. Some national security work cannot be described in detail, but secrecy does not eliminate the need for lawful authority, internal review and meaningful accountability. The use of AI in areas that affect civil liberties or the use of force demands especially careful scrutiny because errors can have consequences that are difficult to reverse.
OpenAI has not announced a new programme as part of Driscoll’s appointment, so the role’s impact will be clearer only when the company describes concrete work or agencies disclose deployments. Useful indicators will include the type of systems being evaluated, the safeguards attached to them, the evidence used to validate performance and whether the company can explain how its technology fits within existing oversight frameworks.
For now, the appointment signals that OpenAI wants a more direct understanding of government and defence requirements as AI moves into higher-stakes settings. Driscoll’s experience may help the company navigate those requirements, but the announcement is a staffing decision, not proof that a new capability is ready for use. The practical test will be whether future deployments are useful, secure, accountable and limited to clearly defined purposes.