There’s a dangerous myth that AI will make war “cleaner” or more “surgical.” This kind of talk completely misses the point and gets in the way of serious discussions about how we’re going to control these systems. As AI gets baked into military and intelligence operations, the real conversation is about governance, who decides what data gets collected on civilians, how an algorithm is allowed to interpret that data, and who’s accountable for the results. If we don’t have clear answers, we risk automated systems making irreversible decisions, like mistaking a wedding convoy for an enemy column, leading to truly catastrophic failures.
Key Takeaways
- Any AI used in a conflict zone has to come with a complete “chain of custody” for its data, meaning a clear, unalterable log showing where every piece of information came from, how it was analyzed, and what actions were based on it.
- The Geneva Conventions and other parts of international humanitarian law absolutely apply to AI in warfare. This means any system must be built to respect the principles of distinction, proportionality, and precaution when planning or executing attacks.
- Global South nations are actively defining AI rules, not just waiting for the West. The African Union’s AI Strategy, for example, insists on data sovereignty, and the ASEAN AI Guidebook focuses on local values, providing essential counterweights to a purely Western-centric view.
- You can fight algorithmic bias in conflict AI. For instance, if you’re building a system to identify military vehicles, you need to train it on images from local experts in that specific desert or jungle, not just generic stock photos, and have an independent team constantly review its performance.
- Holding people accountable for an AI’s mistake is possible if you build the right legal structures ahead of time. This could mean things like creating international agreements that assign clear responsibility to the software developer, the field operator, or the commanding officer who signed off on its use.
Myth 1: AI Data Governance in Conflicts is a Purely Technical Challenge
Too many people think that AI data management in a warzone is just about having good encryption and secure servers. That’s part of it, but it’s like saying building a hospital is just about bricks and mortar. The core of AI data governance in conflict is a messy, human problem of politics and ethics. The tech is just a set of tools to be used within that context.
The data itself represents human lives, movements, and vulnerabilities. Decisions about what information to even collect, let alone how to analyze it and share it, are tangled up in human rights, international law, and raw power dynamics. Take satellite imagery fed into an AI for targeting. The algorithm might spot military assets, but what if it’s right next to a hospital or a school? The final human decision rests on how that data was governed, classified, and presented, all of which has to respect the principles of distinction and proportionality under international law. As a UNIDIR report put it, “data governance for AI in armed conflict is less about the ‘how’ of data management and more about the ‘who’ and ‘why’ of data use, involving complex ethical and legal considerations” (UNIDIR Primer Series: Artificial Intelligence and Armed Conflict).
On top of that, you can’t get meaningful consent for data collection in a refugee camp or a city under siege. So what are the ethics of collecting biometric data from displaced people, even if you have the technical ability? The potential for that data to be misused later is enormous. You need strong policies, independent oversight bodies (think international observers with real teeth), and clear rules for accountability that go far beyond any technical spec sheet.
Myth 2: Existing International Law Doesn’t Apply to AI in Conflict
You often hear that AI is moving so fast that it’s creating a legal “wild west” in warfare. That’s a dangerous idea, and it’s wrong. International humanitarian law (IHL) applies to *all* new weapons and tactics, and that includes systems driven by AI. The bedrock principles of distinction, proportionality, and precaution don’t just disappear because a computer is involved.
The principle of distinction requires fighters to tell the difference between soldiers and civilians, and between military bases and civilian homes. An AI system used for surveillance or targeting has to be designed and used in a way that helps humans make that distinction. If your AI-powered drone can’t reliably tell a farmer with a hoe from a soldier with a rifle, then using it is probably a war crime. The International Committee of the Red Cross (ICRC) has been very clear on this, stating that “new technologies of warfare, including autonomous weapon systems, must comply with international humanitarian law” (ICRC: New technologies of warfare and international humanitarian law).
Likewise, proportionality says that any expected harm to civilians can’t be excessive compared to the direct military gain. A human commander must always be in a position to make that judgment call, even if an AI recommends a target. The principle of precaution in attack demands that commanders take every feasible step to minimize civilian harm, which includes things like rigorously testing an AI’s reliability and ensuring there’s a strong human review process before pulling any triggers.
The real challenge is interpreting and enforcing these laws with complex AI. Figuring out who’s to blame for a violation committed by an autonomous system, the programmer, the operator, the commander, requires a careful look at the chain of responsibility. These are extensions of existing principles to evolving technologies. For a look at how legal systems are trying to catch up with AI in other areas, you can check out this piece on the AI Copyright Act of 2025.
Myth 3: Western Nations Alone Will Dictate Global AI Governance Norms
It’s easy to assume that the rules for AI in conflict will be written in Washington D.C. and Brussels and handed down to the rest of the world. That view completely ignores the real work being done in other regions. Effective AI governance has to be a global conversation, and developing nations are increasingly taking the lead.
We’re seeing this happen right now. The African Union is developing its own Continental AI Strategy, which is built around human-centric design, inclusion, and data sovereignty. It’s a direct response to the specific security challenges and opportunities on the continent. In another example, the Association of Southeast Asian Nations (ASEAN) has published its own AI Governance and Ethics Guidebook, which stresses responsible development that fits with local cultural values and data protection norms.
These initiatives proactively shape the global discourse to make sure AI governance isn’t just a reflection of a few powerful countries’ interests. Nations in the Global South often have firsthand experience with how new technologies affect humanitarian crises, giving them an invaluable perspective on everything from ethical data collection to algorithmic bias. For any global AI governance framework to have legitimacy and actually work on the ground, it has to be built with these diverse viewpoints from the start.
Myth 4: Algorithmic Bias is an Unavoidable Consequence in Conflict AI
Saying that algorithmic bias is an unfixable problem in conflict AI is a lazy and dangerous form of fatalism. The challenges are real, but bias isn’t some magical property of AI. It’s a product of human choices, and it can be fought with deliberate design, tough testing, and constant oversight.
Bias almost always starts with the training data. If an AI is trained to identify threats using data mostly from one region or one demographic, it’s going to be less accurate when you deploy it somewhere else. In a conflict zone, a surveillance system trained on urban European cities might be useless or, worse, dangerously wrong in a dense jungle. An error could lead directly to misidentification and a wrongful attack, which has obvious life-or-death implications.
First, you have to build diverse and representative data sets. This is non-negotiable. It means doing the hard work of collecting and curating data that reflects the actual environment where the system will be used. Second, you need independent ethical reviews at every single stage. These review boards must include ethicists, human rights lawyers, and, most importantly, people from the local communities who can spot potential biases that the developers would never see. Third, continuous monitoring and evaluation in the field is essential to catch biases that only show up under real-world pressure and to update the algorithms. The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems has published a lot of good material on Ethically Aligned Design that gets into the weeds on this.
Myth 5: Accountability for AI Failures in Conflict is Impossible
The “you can’t put an algorithm in jail” argument often leads people to think that holding anyone accountable for an AI’s mistakes in war is a hopeless task. This view seriously underestimates how legal and ethical frameworks can adapt. Accountability for AI failures in conflict, while definitely challenging, is both feasible and necessary.
Think it through with an example. An AI-powered drone gives bad intelligence that leads to civilian deaths. Who’s on the hook? Was the algorithm’s logic flawed from the start? Was it fed bad data? Did a human operator ignore a warning or fail to use the system properly? Or did a commander make a reckless decision based on the AI’s output without seeking confirmation? Accountability means being able to answer those questions.
Legal frameworks are already evolving to create this clarity. Many countries are looking at “human-in-the-loop” or “human-on-the-loop” mandates to make sure a person is always responsible for critical decisions. Existing principles of command responsibility can also apply. The key to all of this is transparent documentation. You need logs of the AI’s design, its decision-making criteria, and its operational history. These audit trails are what allow investigators to trace a failure from the tragic outcome back to its source. The European Commission’s AI Act, for example, sets out rules for human oversight and conformity assessments for high-risk AI, offering a model for how this kind of accountability can be put into law.
In the end, making AI accountable means getting lawyers, ethicists, engineers, and policymakers in a room to build clear frameworks that ensure there’s justice when things go wrong in an increasingly AI-driven world.
If we want to use AI responsibly in global conflicts, we have to get past these myths. The real work is about applying established laws to new tools, demanding global input, actively fighting bias, and designing clear chains of accountability from the very beginning. These aren’t optional extras. They’re the only way to manage the deep risks of this technology.
What is the primary role of data governance in ethical AI for conflict zones?
Data governance is about setting the rules of the road for how information is collected, used, and stored by AI in a conflict. Its main job is to make sure every action sticks to ethical principles and international law, protecting human rights by ensuring you know where the data came from, can vouch for its integrity, and can control who sees it.
How does international humanitarian law (IHL) apply to AI in armed conflict?
IHL applies to AI the same way it applies to a rifle or a tank. It requires that any AI system used in war must obey the core principles of distinction (don’t target civilians), proportionality (don’t cause excessive civilian harm), and precaution (take steps to minimize civilian casualties). The people who build and use these systems are responsible for making sure they can follow these rules.
Can algorithmic bias in AI systems used in conflict be completely eliminated?
Probably not completely, but calling it unavoidable is an excuse. You can seriously reduce algorithmic bias by using high-quality, diverse training data, bringing in independent ethical review teams, and constantly monitoring the system’s performance in the real world with input from local experts. It’s a constant process, not a one-time fix.
Who is accountable when an AI system causes harm in a conflict setting?
Accountability is a chain that can include the developers, the organization that deployed the system, the individual operator, and the commander who authorized its use. Figuring out who’s at fault requires clear legal frameworks which are currently being developed. The key is having transparent logs and human oversight so you can trace the error back to its source.
What role do non-Western nations play in shaping global AI governance for conflict?
They play a huge role by preventing a small group of tech-heavy countries from setting all the rules. Groups like the African Union and ASEAN are creating their own AI strategies, like the AU’s AI Strategy and the ASEAN AI Guidebook, that focus on things like data sovereignty and local values. Their involvement is essential for creating global rules that are seen as legitimate and work for everyone.