A detailed proposal for governing AI contracts with government agencies lays out a new approach to balancing national security work with ethical safeguards. Rather than banning defense collaborations outright, the framework establishes clear limits on how AI can be used in targeting and surveillance while creating an internal oversight process intended to keep company leadership accountable.
The proposal argues that companies building advanced AI systems should adopt enforceable internal standards before lawmakers establish binding regulations. It positions transparency, documented due diligence, and technical enforcement as the foundation of responsible government AI partnerships.
What the framework proposes
At the heart of the proposal are two core standards.
The first would prohibit company AI from being used in systems that select and engage targets without appropriate human control over every engagement decision. The requirement extends beyond complete weapons systems and also covers AI components that contribute to targeting pipelines.
The second standard limits AI driven surveillance. It would prevent government customers from using the company's AI to transform large datasets into individualized intelligence on people who have not already become legitimate subjects of investigation based on specific facts. The proposal also bars AI generated outputs from serving as the sole basis for opening investigations and forbids profiling based only on demographic characteristics, religion, ethnicity, nationality, or political expression.
The framework includes limited exceptions for missile defense, aggregate statistical analysis, logistics, translation, and other lower risk military applications.
Oversight without a corporate veto
Instead of granting an ethics board the authority to block contracts, the proposal creates a seven member Defense AI Review Body with an advisory role.
Its responsibilities include:
- Reviewing contracts involving targeting or surveillance before execution.
- Auditing lower risk contracts after they are signed.
- Issuing written compliance findings.
- Publishing an annual transparency report for employees.
- Recording every instance where executives override a non compliance finding.
Leadership would retain the final decision on contracts. However, any disagreement between executives and the review body would become part of a permanent internal record. Repeated overrides would be disclosed in yearly transparency reports, making it difficult to quietly weaken the oversight process.
Technical controls remain central
The proposal argues that governance cannot rely only on policy documents. It recommends deployment rules based on how much technical control the AI provider retains.
Air gapped deployments would be prohibited for targeting, biometric identification, or individualized surveillance. Where disconnected deployments are allowed, only narrowly scoped models would be supplied. Those models would undergo capability removal techniques intended to make repurposing impractical.
Legal and business motivations
The proposal presents the framework as more than an ethics initiative. It also frames it as a legal risk management strategy.
According to the document, companies providing AI for military or surveillance purposes without documented compliance procedures could face negligence claims if their systems contribute to unlawful harm. A structured review process would create evidence that the company exercised reasonable care before entering sensitive contracts.
The framework also recommends placing review body deliberations under attorney client privilege. That would allow the company to preserve confidential legal assessments while selectively disclosing its compliance process if needed during litigation.
Why the proposal matters
Governments around the world are expanding their use of generative AI, computer vision, and autonomous systems. Yet comprehensive legislation governing military AI and AI assisted surveillance remains limited in many jurisdictions.
This proposal attempts to fill that gap through corporate governance instead of waiting for regulation. It also recognizes that AI companies increasingly face pressure from employees, investors, governments, and the public to explain how their technology is used once it leaves their control.
If adopted by a major AI provider, the framework could become an early reference point for future industry standards or legislation. It also reflects a broader shift in AI governance, where technical safeguards, transparency, and documented accountability are becoming as important as model performance itself.
What comes next
The proposal envisions that its standards would remain in place unless future legislation establishes stronger and independently enforceable protections. Until then, it argues that AI companies should adopt clear internal rules before expanding government partnerships, setting expectations that could shape future contracts across the defense and public sector ecosystem.