OpenAI has hired Dan Driscoll, who served as the 26th Secretary of the U.S. Army until early September, as its Senior Counselor for National Security Modernization. The company announced the newly created role on November 2, saying Driscoll will work with senior leadership and outside partners on how advanced AI can strengthen national and cyber security, “with a particular focus on working with U.S. allies to advance its responsible use.”
The Wall Street Journal, which reported the hire, framed it as part of OpenAI’s push to win government contracts and shape military AI policy. OpenAI’s own announcement does not mention contracts. It says Driscoll will advise public and private sector partners on “modernizing policy and strategy choices,” work with governments, industry, academia and allied institutions on what the company calls “democratic AI,” and help OpenAI’s internal teams understand the operational needs of national security institutions.
For privacy and compliance teams, the personnel move is less important than what it signals. The largest consumer AI company is steadily becoming a defense and government contractor, and the data rules that govern government AI work are very different from the ones that govern a chatbot subscription. Companies that sell to agencies, work as subcontractors, or use the same AI platforms for commercial and public sector clients should be paying attention.
Who Driscoll is and how he left the Army
Driscoll, an Iraq war veteran and Yale Law School graduate who worked in venture capital and private equity, led the Army during the second Trump administration. According to DefenseScoop, he launched the Army Transformation Initiative, the service’s drone and counter-drone marketplaces, and Operation Jailbreak, an effort to get disparate Army systems to connect. He was also drawn into U.S.-brokered talks on ending the war in Ukraine, which Reuters called a rare diplomatic task for an Army secretary.
His departure followed months of reported tension with Defense Secretary Pete Hegseth, including over Hegseth’s removal of Gen. Randy George as Army chief of staff. Driscoll submitted his resignation on August 31, and in a social media post on September 1 said the following day would be his last full day in the role. Reports that describe him as resigning “in August” or “in September” are both partly right: the resignation was submitted in August and took effect in early September. OpenAI’s announcement also notes that he served as acting director of the Bureau of Alcohol, Tobacco, Firearms and Explosives.
In a statement released by OpenAI, Driscoll said national security requirements are “changing faster than at any time in the past fifty years,” and that AI can help “by cutting through paperwork, making sense of more information, and getting the right answer to the right person faster.”
OpenAI’s government business, by the record
The hire lands on top of a defense relationship that has grown quickly. The public record so far includes:
- June 2025: the Pentagon’s Chief Digital and AI Office awarded OpenAI Public Sector a $200 million prototype Other Transaction agreement to develop frontier AI capabilities for “warfighting and enterprise domains,” as DefenseScoop recounted.
- February 28, 2026: OpenAI published the terms of an agreement to deploy its models in classified Pentagon environments, reached after the department’s dispute with Anthropic. OpenAI said the deployment is cloud-only, runs OpenAI’s own safety stack, keeps cleared OpenAI personnel in the loop, and is bound by three red lines: no mass domestic surveillance, no directing autonomous weapons, and no high-stakes automated decisions.
- March 2, 2026: OpenAI added contract language stating the system “shall not be intentionally used for domestic surveillance of U.S. persons and nationals,” including through “commercially acquired personal or identifiable information,” and said the department affirmed the services would not be used by its intelligence agencies such as the NSA without a new agreement.
- August 2026: officials said users of GenAI.mil, the Pentagon’s enterprise generative AI platform, could use approved versions of OpenAI’s ChatGPT models.
- Ongoing: OpenAI’s Daybreak program, which the Driscoll announcement cites, offers subsidized AI cyber-defense access to essential-service operators such as utilities, community banks and state and local governments.
Critics, including writers at The Atlantic and advocacy groups, have argued that words like “intentionally” and “deliberate” in the surveillance clause leave room for incidental collection, and the full contract has not been published. OpenAI says its layered approach is stronger than relying on usage policies alone. Neither side’s claim can be fully tested from the outside, which is itself a governance point: the public can see the summary of the safeguards, not the controls.
Why a defense hire is a data governance story
When an AI provider serves both consumers and the national security apparatus, three questions follow for anyone whose data or contracts touch both worlds.
Whose data trains what?
Federal procurement rules now speak to this directly. The White House Office of Management and Budget’s April 2025 memo M-25-22, “Driving Efficient Acquisition of Artificial Intelligence in Government,” directs agencies to include contract terms that permanently prohibit vendors from using non-public agency inputs and outputs to train publicly or commercially available AI models unless the agency explicitly consents. That requirement applies to AI contracts under solicitations issued on or after October 1, 2025, and to covered contracts renewed or extended after that date. Contractors and subcontractors in the AI supply chain should expect those terms to flow down to them.
Where commercial data meets government use
The most privacy-relevant line in OpenAI’s Pentagon contract is the reference to “commercially acquired personal or identifiable information.” That is the data broker market: location, browsing and identity data bought rather than collected under a warrant. Its inclusion confirms that regulators, contractors and the AI labs themselves see AI as a force multiplier for that data. Companies that sell or share personal data commercially should assume it can end up in government analytics pipelines unless their contracts and state data broker laws say otherwise. Public concern about this is not abstract; we covered Americans’ views on government data use in our analysis of public distrust in government data collection, and the fallout of federal data access fights in our report on calls to tighten government privacy laws.
Who is in the room when rules get written
Former senior officials joining AI companies is legal and common. Federal post-employment rules, including 18 U.S.C. § 207, generally restrict former senior officials for a period after leaving government from contacting their former agency on someone else’s behalf to seek official action, among other limits. Those rules govern lobbying contacts; they do not settle the broader policy question of how much influence AI vendors should have over how the military adopts AI. Corporate buyers should notice that the companies shaping government AI norms are the same ones selling them enterprise tools.
Five checks for companies in the government AI supply chain
Whether you are a prime contractor, a subcontractor, a software vendor that embeds a frontier model, or a business that simply uses the same AI platforms as government customers, these are practical steps to take now:
- Map which AI providers touch government data. Record every model provider, hosting environment and subprocessor in your public sector work, separately from your commercial deployments. Mixing tenants is how restricted data ends up in the wrong place.
- Check your contracts for M-25-22-style terms. Look for no-training clauses, data ownership language, portability and deletion obligations, and make sure your own upstream AI vendor contracts let you meet them.
- Separate commercial personal data from government work. If your company holds consumer data, confirm whether any of it can be sold, licensed or shared with public sector clients, and whether state data broker or privacy laws restrict that.
- Review AI tool use by staff on government accounts. Several governments have restricted public chatbots in their own offices; Sweden’s Government Offices, for instance, blocked ChatGPT, Claude and Copilot, as covered in our report on Sweden’s AI tool block. Your employees working on agency contracts may face similar limits.
- Prepare for scrutiny of national security exposure. Boards and customers increasingly ask about national security compliance, from export controls to foreign adversary data rules. Our coverage of companies unprepared for national security compliance risks shows how wide that gap still is.
Captain Compliance’s platform includes AI governance, data mapping and vendor risk tools that can help record which AI systems and providers process which categories of data, and keep evidence of what was approved. Records like that support contract reviews and audits; they do not replace legal advice on federal acquisition rules or a contracting officer’s interpretation.
What the hire does not tell us
It is worth being clear about what is known and what is not:
- OpenAI has not announced any new government contract tied to Driscoll’s arrival.
- The WSJ’s characterization that he will help secure government contracts is the newspaper’s reporting; OpenAI’s release describes an advisory and partnership role.
- The full text of OpenAI’s classified deployment contract is not public, so outside verification of the surveillance and autonomy safeguards is limited to what the company has published.
- How OpenAI separates consumer, enterprise and government environments in practice has been described by the company at a high level only.
OpenAI has drawn regulatory attention on the consumer side for years, and AI governance failures tend to be expensive when they surface; see the most costly AI governance failures of the past two years. Its earlier move to restrict access to an advanced cybersecurity model, discussed in our look at OpenAI’s guarded cyber capability, shows the company already sorting customers by trust level. A senior national security hire suggests that sorting will become more formal.
Frequently asked questions
What is Dan Driscoll’s role at OpenAI?
He is Senior Counselor for National Security Modernization, a new position announced November 2, 2026. OpenAI says he will advise leadership and external partners on using AI for national and cyber security, with a focus on U.S. allies, and help internal teams understand national security institutions’ needs.
When did Driscoll leave the Army?
He submitted his resignation on August 31, 2026, and said on September 1 that the next day would be his last full day as secretary, so he left in early September.
Can the Pentagon use OpenAI models to watch Americans?
OpenAI says its contract bars intentional domestic surveillance of U.S. persons and nationals, including through commercially acquired personal data, and that use by intelligence agencies such as the NSA would require a new agreement. Critics say the wording leaves gaps. The full contract has not been released.
Does this change anything for ordinary ChatGPT or enterprise users?
Not directly. Government deployments run under separate contracts and environments. The indirect effect is that the vendor’s priorities, staffing and policy positions increasingly reflect government customers, which enterprise buyers should factor into their vendor reviews.
Do federal AI procurement rules apply to subcontractors?
OMB’s M-25-22 directs agencies on what to put in their AI contracts. Whether specific terms flow down to your company depends on your prime contract and subcontract language, so check both.
The counselor and the contract
OpenAI’s announcement leans heavily on the words “responsible” and “democratic.” Those words will be tested not in press releases but in contract clauses, audit logs and the data flows behind them. A former Army secretary can open doors in Washington and allied capitals. The open question for privacy teams is whether the safeguards OpenAI has published for its government work will become something customers and the public can verify, or remain a summary they are asked to trust.
