Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsOpenAI said on February 28, 2026, that it had reached an agreement with the Pentagon—called the “Department of War” in the company’s announcement—to deploy advanced AI systems in classified environments. The company says the arrangement preserves a company-controlled safety stack, requires cloud-only deployment and cleared OpenAI personnel, and restricts certain uses involving domestic surveillance, autonomous weapons and high-stakes automated decisions.
That does not mean the Pentagon agreed to stop using AI for military purposes, or that the safeguards have been independently verified. The public record is primarily OpenAI’s description of the agreement. Its full text, technical implementation, audit process and enforcement history have not been established in the available sources.
As an Amazon Associate I earn from qualifying purchases.
What OpenAI announced
Sam Altman announced the agreement publicly on February 28, according to TechCrunch. OpenAI’s more detailed official account says the deal will allow its advanced AI systems to operate in classified environments.
Free tools Windows power users keep installed
One-click scans. No signup required.
OpenAI said the agreement is intended to address a practical concern: the U.S. military wants access to powerful AI systems as other countries integrate AI into military and national-security operations. At the same time, the company said it wanted to preserve restrictions it considers essential and reduce the confrontation between the Pentagon and AI companies.
OpenAI also asked that comparable terms be made available to other AI laboratories. That request matters because different rules for different vendors could push companies to compete by offering fewer restrictions rather than better safeguards.
What “technical safeguards” means
The phrase does not describe one feature or a single switch. OpenAI describes a layered control model combining technical architecture, personnel, contractual restrictions and existing legal or policy requirements.
- Company-controlled safety stack: OpenAI says it will retain control over the safety systems protecting the deployment.
- Safety-trained models: The company says it will not provide “guardrails-off” or non-safety-trained models.
- Cloud-only deployment: OpenAI describes the arrangement as cloud-only rather than an unrestricted model deployment to edge devices.
- Classifiers: OpenAI says it can independently verify and update classifiers used to detect or restrict prohibited activity.
- Cleared personnel: Cleared OpenAI engineers, along with safety and alignment researchers, are described as remaining involved with the deployment.
These safeguards should not be treated as interchangeable. A classifier is a technical control. A restriction written into a contract is a contractual control. A requirement for human approval is an oversight rule. Constitutional, statutory and Department policies are legal or policy constraints. Each can reinforce the others, but none automatically guarantees that a prohibited use cannot occur.
Recommended Free Tools
The three stated red lines
1. Mass domestic surveillance
OpenAI says its technology may not be used for mass domestic surveillance. Its original description referred to the Fourth Amendment, the National Security Act of 1947, the Foreign Intelligence Surveillance Act of 1978, Executive Order 12333 and applicable Department directives requiring a defined foreign-intelligence purpose.
OpenAI’s account also said the system could not be used for unconstrained monitoring of U.S. persons’ private information or for domestic law-enforcement activities except where legally permitted.
Rank #2
On March 2, OpenAI added more explicit language. The company said the restriction covers intentional tracking, surveillance or monitoring of U.S. persons or U.S. nationals, including surveillance conducted through commercially acquired personal or personally identifiable information.
The update also said that use by Department of War intelligence agencies such as the National Security Agency would require a new agreement. This is an important clarification, but it does not amount to a general ban on intelligence work, foreign-intelligence collection or analysis of government-supplied data. The precise legal and operational boundaries would depend on the agreement, applicable law and how the data is used.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →2. Independent direction of autonomous weapons
OpenAI says the contract prohibits its AI system from independently directing autonomous weapons when law, regulation or Department policy requires human control. The company also cited DoD Directive 3000.09, dated January 25, 2023, which requires rigorous verification, validation and testing for AI used in autonomous and semi-autonomous systems before deployment in realistic environments.
This is narrower than a ban on all weapons-related AI. The described language does not prohibit military planning, intelligence analysis, targeting support or other applications merely because they relate to defense. It addresses independent direction of autonomous weapons in circumstances where existing rules require human control.
“Human control” also needs to mean more than a person formally approving a machine-generated recommendation. Meaningful oversight would normally require the operator to understand the relevant information, have enough time to assess it, possess authority to reject the action and be able to intervene effectively. The public description does not establish how those conditions will be measured in live operations.
Rank #3
3. High-stakes automated decisions
OpenAI says its technology may not be used for high-stakes automated decisions, giving “social credit” as an example. The available announcement does not provide a complete public list of covered decisions.
That leaves important questions. Would the restriction apply to automated decisions affecting personnel, detention, eligibility, benefits or access to services? Would it cover systems that do not make the final decision but rank people or cases for a human official? The answer may depend on the contract’s definitions and operational rules, which have not been publicly detailed in the sources available here.
What the agreement does not prohibit
OpenAI did not describe the arrangement as a ban on military AI. The company says the systems can be used for “all lawful purposes,” subject to applicable law, operational requirements and safety and oversight protocols.
That broad phrase is central to understanding the deal. The agreement appears to permit extensive military and national-security use while placing limits on specified categories. It would therefore be inaccurate to summarize the announcement as either “OpenAI gave the Pentagon unrestricted AI” or “OpenAI banned military use of its models.” Neither description matches the company’s public account.
Why the Anthropic dispute shaped the deal
The announcement followed a public standoff between the Pentagon and Anthropic. TechCrunch reported that the Pentagon sought authorization for “all lawful purposes,” while Anthropic sought restrictions involving mass domestic surveillance and fully autonomous weapons.
Anthropic chief executive Dario Amodei argued that, in a narrow set of cases, AI use could undermine democratic values rather than protect them. Defense Secretary Pete Hegseth, according to the report, said Anthropic was attempting to obtain veto power over U.S. military operational decisions. President Donald Trump directed federal agencies to stop using Anthropic’s products after a six-month phase-out period, the report said.
Those statements describe the political dispute and should not be treated as a settled characterization of Anthropic’s position. Anthropic’s stance was not that it opposed all military work; the disagreement concerned particular uses and the company’s ability to enforce limits on them.
OpenAI presented its agreement as addressing the same two central issues—domestic mass surveillance and autonomous weapons—while adding a restriction on high-stakes automated decisions. The arrangements should not be assumed to be identical, however. The public information does not establish that OpenAI accepted every term Anthropic sought or that the two companies’ contracts use the same definitions and enforcement mechanisms.
Technical controls versus promises and rules
OpenAI’s description combines several layers that work differently:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Layer | What it can do | What it cannot prove by itself |
|---|---|---|
| Model refusals and safety training | Discourage or block certain requests during interaction. | Guarantee correct behavior in every novel, adversarial or operational context. |
| Classifiers | Detect patterns associated with restricted content or uses. | Eliminate false negatives, false positives, distribution shift or attempts to disguise intent. |
| Cloud-only architecture | May give OpenAI greater ability to control access, update safeguards and monitor use than an uncontrolled local deployment. | Prevent downstream systems from acting on permitted outputs or resolve every integration risk. |
| Human review | Keep a person involved in specified decisions. | Show that the person had sufficient information, time, authority and ability to intervene. |
| Contract terms | Create obligations and a basis for challenging prohibited use. | Guarantee immediate detection or enforcement, especially in classified workflows. |
| Law and Department policy | Set external limits on surveillance, weapons and other activities. | Make compliance automatic without testing, supervision and accountability. |
Cloud-only deployment may reduce some risks associated with copying a model onto uncontrolled devices, but it is not a complete answer to military AI risk. A cloud service can still be connected to downstream tools, data systems or operational processes. The public account does not explain every possible integration.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Questions the public record does not answer
OpenAI says it could terminate the contract if the government violated its terms. But the practical value of that remedy depends on details that have not been publicly established, including:
- the full contract text and its definitions;
- the specific model or product being deployed;
- the deployment timeline and classified environment;
- the cloud provider and system architecture;
- logging, access-control and audit requirements;
- incident reporting and breach-notification procedures;
- independent oversight or verification;
- the process for investigating suspected misuse;
- the consequences for agencies, contractors or users who violate the terms; and
- whether the Pentagon has independently confirmed every part of OpenAI’s description.
The available sources establish no public breach or enforcement case. They also do not establish that OpenAI has independently audited Pentagon compliance.
How the safeguards could be tested in practice
The hardest cases are likely to involve boundaries rather than obvious prohibited requests:
- Formal approval: A person approves a model recommendation without enough time, context or authority to exercise genuine judgment.
- Indirect weapons support: The model does not directly control a weapon but prioritizes targets, fuses intelligence or recommends a mission action.
- Data repurposing: Information collected for foreign-intelligence purposes is later used for domestic policing or surveillance.
- Commercially obtained data: Location, identity or behavioral data is analyzed without the operator describing the activity as surveillance.
- Model updates: A new model or classifier changes behavior inside a classified environment.
- Emergency operations: Urgent military needs create pressure to bypass ordinary review or access controls.
- Agency and contractor access: A system approved for one Pentagon component becomes accessible to an intelligence agency or contractor.
- High-stakes administration: The model influences detention, personnel, eligibility or benefits decisions without formally making the final decision.
These are risk scenarios, not reported incidents. They show why the quality of the agreement cannot be judged solely by whether it contains the words “human control” or “no surveillance.” Implementation, monitoring and enforceability determine whether the safeguards have practical effect.
Why the announcement matters
The deal is significant because it presents a possible middle path between unrestricted military access and a complete refusal to serve defense customers. OpenAI says the military can use advanced systems for lawful purposes while the company retains control over technical safeguards and draws contractual red lines.
It also raises a broader governance question: can a private AI company maintain meaningful control over its systems after placing them inside a classified defense environment? The answer depends less on the announcement’s headline than on access controls, auditability, model-update procedures, human authority, incident reporting and the company’s willingness and ability to enforce the contract.
For now, the most accurate description is limited but consequential: OpenAI says it secured a layered safeguard regime for classified Pentagon deployments. The public materials do not yet show how independently verifiable or operationally enforceable that regime will be.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
