In March 2026, India's Defence Ministry told a parliamentary panel that DRDO and defence public sector undertakings were actively researching lethal autonomous weapon systems. The ministry also acknowledged the limits of the technology, stating that AI and machine-learning techniques are "not amenable to verified decision-making" and could lead to unintended outcomes. Its preferred approach was semi-automatic control, with AI supporting rather than replacing human decisions.

That distinction matters. India is pursuing the capabilities AI can offer while recognising that some decisions cannot be handed over to a machine without careful consideration of the consequences.

The rest of the world is still working through the same question. In September 2026, a Swedish loitering-munition demonstration made headlines after reports described a drone selecting its own target. The company later clarified that a human had verified the target and that the strike was simulated. The same demonstration was therefore discussed as an example of a machine choosing whom to strike and as an engagement that still involved human judgment.

The difference between those two accounts is significant. It shows how easily a discussion about autonomy can become a debate about labels rather than an examination of what a system actually does. A machine that analyses sensor data, identifies objects or recommends a course of action is not necessarily making the same decisions as one authorised to engage a target. Understanding where those boundaries lie is essential to understanding both a system's capabilities and its risks.

This is also why the international debate matters. In September 2026, the 128 states party to the Convention on Certain Conventional Weapons agreed on a draft report affirming the importance of human control and judgment over weapons with autonomous functions. A review conference in Geneva, scheduled for 16–20 November, will determine what happens next.

India has taken a distinct position in these discussions. It maintains that international humanitarian law applies in full, while arguing that the debate should not stigmatise the underlying technology. It has also favoured a consensus-based outcome within the CCW rather than a new binding treaty.

Accountability starts with how systems are built

Whatever direction the international rules take, the practical question for anyone building autonomous defence systems in India remains the same: when a machine makes a decision that leads to harm, who is responsible?

The answer cannot simply be the operator sitting in front of a screen. A system's behaviour reflects decisions made long before it reaches the field. Its designers determine what it is optimised to detect, which signals it prioritises, how it handles uncertainty and when it hands control back to a person. Its manufacturer determines how it is tested, updated and maintained. Its users determine how it is deployed and what authority it is given. Accountability requires understanding how these responsibilities fit together, rather than treating the machine as an independent actor that can absorb the blame.

This principle is increasingly being recognised within the defence industry itself. Shield AI co-founder Ryan Tseng recently compared accountability in autonomous systems to aviation, arguing that if you designed the aircraft, you are accountable for what it does. The analogy is useful because a complex system does not become exempt from responsibility simply because software is involved in its operation. If a system's design contributes to a harmful outcome, saying "the algorithm did it" explains very little.

Accountability also depends on who controls the technology. A force cannot fully inspect or independently assess a system if critical parts of its logic are inaccessible, or if access depends on a foreign vendor's licences and export restrictions. At the India-AI Impact Summit in 2026, DRDO Director General Chandrika Kaushik made this point directly, saying that Indian defence cannot afford to depend on AI models from abroad. Developing and maintaining systems in India can give Indian institutions greater control over their underlying logic, updates and evaluation. That does not automatically make a system safe or accountable, but it makes meaningful oversight more achievable.

India already has a framework for evaluating trustworthy defence AI. In October 2024, DRDO launched the Evaluating Trustworthy AI (ETAI) framework, a risk-based assessment framework for AI used in critical defence operations. It rests on five principles: reliability and robustness, safety and security, transparency, fairness, and privacy. The challenge is ensuring these principles influence how systems are engineered and tested, rather than remaining statements of intent.

For mission software, this begins with making human authority explicit. The system must distinguish between what it can do independently, what it can recommend and what requires an operator's decision. Operators also need visibility into its behaviour: they should be able to understand what the system has done, where it is acting autonomously and when human intervention is required. Finally, decisions need to be traceable. Recording the information available to the system, the recommendations it produced and the approvals given by operators makes it possible to reconstruct events and review the reasoning behind a decision.

These are practical requirements, not just ethical aspirations. Indian forces need to be able to explain the behaviour of the systems they deploy to their own chain of command. Evaluators need to assess whether those systems meet the relevant requirements, and commanders need confidence that they understand the limits of the capabilities available to them. A system may perform impressively in a demonstration, but if its decisions cannot be explained, its behaviour cannot be reviewed or its boundaries cannot be established, those gaps can become obstacles to deployment. In defence, trust is part of what makes the technology usable.

A mosaic illustration of a robotic finger and a human finger almost touching

Making human control practical

None of this means autonomy has little value. Used well, it can reduce the workload on operators and help them make better decisions with the information available. Consider the Army's Ashni drone platoons, each of which fields around ten types of drones. An operator responsible for a complex mission may need to monitor multiple video feeds, track the position and battery levels of different assets, interpret incoming sensor information and maintain an understanding of what is happening across the wider area. Expecting one person to process all of this simultaneously is unrealistic. Important details can get lost, and time spent monitoring routine activity is time that cannot be spent assessing situations that genuinely require judgment.

Software can help by bringing sensor data together, flagging changes, tracking assets, coordinating platforms and highlighting the information that needs attention. Instead of forcing operators to piece together a mission from disconnected screens, it can give them a clearer picture of what is happening across the systems under their control. The point is not to remove people from the process, but to reduce the burden of gathering and processing information so they can focus on the decisions that matter.

Where Driif stands

These principles shape how we believe mission software should be built. Driif's Mission OS is designed to coordinate unmanned systems from different manufacturers, helping operators manage platforms as part of a mission rather than as separate tools. Our role is orchestration under human direction, not uncontrolled swarming. That distinction matters because coordinating multiple systems should make human oversight more effective, not make it harder to understand what the machines are doing.

Our position rests on six commitments. Humans stay in command: our software supports human decisions without replacing human authority over a mission. Authority is designed in, so the boundaries between an operator's decisions and automated functions are clear. Every decision is traceable, with records of what the system received, suggested and did so its behaviour can be reviewed. We answer for what we build and accept responsibility for how our software is designed when it contributes to a decision. We believe the code and logic should remain under Indian control, allowing the forces that use them to inspect and audit the technology. And we orchestrate rather than swarm: our role is to coordinate systems under human direction, not to release machines to act on their own.

These are commitments against which our work should be judged, not simply claims about what responsible technology ought to look like. They also reflect a broader view of autonomy. A system is not more useful simply because it makes more decisions independently. Its value depends on whether it helps people understand a situation, coordinate available capabilities and act with the information they need.

The international rules governing autonomous systems are still being debated in Geneva. India's approach is taking shape through DRDO's frameworks, defence policy and the decisions the armed forces make as they bring new systems into service. Builders do not need to wait for every international question to be settled before deciding how their technology should behave. They can make human authority explicit, ensure decisions can be traced and reviewed, retain meaningful control over the software they develop, and accept responsibility for the systems they put into the world.

Autonomy will continue to expand what defence systems can do. The question is whether that progress will be accompanied by equally deliberate thinking about who remains in control and who answers when things go wrong.

Responsible autonomy is not a brake on progress. It is how the future of autonomous missions gets built.

Sources