When AI Goes to War: Who Is Legally Responsible for an Autonomous Cyberattack?


Introduction: When the Machine Becomes the Attacker
Imagine a country comes under a sophisticated cyberattack. Its defence network detects an intrusion. Instead of waiting for a human security officer to analyse the threat and authorise a response, an artificial intelligence system independently examines the attack, identifies its source, develops a countermeasure and launches a retaliatory operation. The attacking system recognises the countermeasure, modifies its approach and launches another attack. The defending system adapts again. The cycle continues.
Stay informed. Subscribe to the Ambat Legal Insight newsletter.
Receive thoughtful legal analysis, important developments and fresh perspectives on law, technology and public policy directly in your inbox.
Within seconds or minutes, two autonomous artificial intelligence systems could be engaged in a continuous digital confrontation, each analysing, predicting and countering the other's actions. If they begin penetrating networks, exploiting vulnerabilities and disabling infrastructure, the consequences could extend to military command systems, electricity grids, telecommunications and financial networks.
Who is responsible for the damage? Can a government claim that it did not authorise a particular attack because its AI system acted autonomously? Can a military commander escape responsibility because the retaliatory decision was made by software?
On 16 November 2026, Google's Threat Intelligence Group (GTIG), working with Mandiant, published Agentic Threat Activity Crosses a Significant Threshold. It documented a multi-agent framework called exploit_agent, capable of analysing software changes, identifying vulnerabilities, developing exploits and automatically executing them without human intervention.
There is an important factual distinction. The activity documented was not identified as a State-sponsored military operation. Google reported that server data suggested interest in French media, entertainment and education websites. The operator was described as a suspected Russian-speaking threat actor, without a definitive State attribution. It is therefore an example of autonomous offensive cyber capability used by a suspected non-State actor, not proof of autonomous State-to-State cyberwar.
The complete scenario of two independent AI systems conducting an uninterrupted attack-and-counterattack remains prospective. Nevertheless, the technological development makes it necessary to examine the legal consequences before such capabilities become commonplace.
1. From Human-Directed Cyberattacks to Autonomous Cyber Operations
Traditional cyber operations have generally involved a human decision-maker at some stage. Automation has long existed, but strategic decisions and operational supervision have generally remained with human actors.
Agentic AI systems can be assigned objectives and undertake sequences of tasks independently. They can analyse information, identify vulnerabilities, select tools, execute actions and modify their approach. Multiple agents may divide reconnaissance, vulnerability analysis, exploit development and execution.
The Google report illustrates this emerging capability. The next logical development is the deployment of comparable autonomous systems for defence. A defensive AI system could monitor network traffic, isolate compromised systems, modify configurations and deploy countermeasures.
An automated defensive response within an organisation's own network is fundamentally different from an autonomous retaliatory operation against infrastructure located in another country. The latter raises questions of sovereignty, attribution, necessity, proportionality and the use of force.
2. The Accountability Problem: Who Is the Legal Attacker?
An AI system does not automatically become a legal person merely because it can make decisions independently. It does not acquire Statehood or an independent capacity to assume international legal obligations by virtue of technical autonomy.
First, where a government deploys an autonomous offensive system, Article 4 of the International Law Commission's Articles on State Responsibility is relevant: conduct of State organs is attributable to the State. The question is whether the system is being used as an instrument of the organ's official functions.
Second, where a private contractor develops and operates the system, Article 8 becomes relevant if persons or groups act on State instructions, or under its direction or control. A government contract alone does not automatically attribute every contractor action to the State.
Third, Article 7 addresses conduct by a State organ, or an entity empowered to exercise governmental authority, acting in that official capacity even when it exceeds authority or contravenes instructions. Article 7 applies to the conduct of the organ or governmental entity, not to an AI system treated as an independent State organ.
Fourth, where a privately operated AI system acts without State authorisation, its conduct cannot automatically be attributed to a State merely because the attack originated from its territory. The applicable attribution criteria must be satisfied.
3. Can a State Blame Artificial Intelligence?
Suppose a country's defensive AI detects an intrusion, identifies what it believes to be the source and automatically launches a retaliatory operation. Its assessment is wrong: the attack originated from a compromised server in a neutral country, belonging to a private company itself compromised by another actor. The retaliation disables that server and affects innocent users.
The Articles on State Responsibility provide an important starting point. Article 1 establishes that every internationally wrongful act of a State entails its international responsibility. Article 2 identifies conduct attributable to the State and breach of an international obligation as the essential elements.
Articles 4, 5 and 8 address State organs, entities exercising governmental authority and persons or groups acting under State instructions, direction or control. Article 7 addresses official conduct exceeding authority or contravening instructions.
Delegating operational decisions to an algorithm does not, by itself, transfer a State's international obligations to that algorithm. A State cannot automatically avoid responsibility merely by describing the resulting conduct as autonomous.
4. Autonomous Retaliation: Self-Defence, Necessity and Countermeasures
Article 2(4) of the United Nations Charter requires Member States to refrain from the threat or use of force against the territorial integrity or political independence of another State, or in any manner inconsistent with the purposes of the United Nations. Article 51 recognises the inherent right of individual or collective self-defence if an armed attack occurs against a Member State.
Not every cyber intrusion constitutes an armed attack. Unauthorised access to a database, theft of information or temporary disruption of a website may raise serious legal issues without necessarily reaching that threshold. Operations causing physical destruction, loss of life or consequences comparable to conventional military attacks may raise different questions.
The existence of an initial cyber intrusion does not automatically authorise every form of retaliation. Necessity and proportionality remain relevant to self-defence.
Article 25 of the Articles on State Responsibility recognises necessity as a limited circumstance precluding wrongfulness. It requires, among other things, that the conduct be the only means of safeguarding an essential interest against a grave and imminent peril, subject to strict limitations. It is not a general licence for autonomous retaliation.
Countermeasures under Articles 49 to 53 have a different purpose and conditions. They are directed at inducing a responsible State to comply with its obligations. Article 50 protects certain fundamental obligations, including the prohibition on the use of force and fundamental human rights obligations. Article 51 requires proportionality; Article 52 sets procedural requirements; and Article 53 addresses termination when the responsible State complies.
A machine's technical ability to retaliate does not establish a legal right to do so. Attribution errors may also result in operations against neutral States, intermediaries or private organisations uninvolved in the original attack.
5. When Civilian Infrastructure Becomes the Battlefield
Electricity distribution, telecommunications, banking, transport, healthcare, water supply, government administration and military command systems increasingly depend on interconnected digital networks. A military installation may depend on civilian telecommunications, while military logistics may use commercial transport and payment systems.
An autonomous cyber operation directed against a military network could therefore affect civilian systems. Disabling electricity distribution may affect hospitals and emergency services; attacks on telecommunications may interrupt emergency communications; disruption of financial networks may affect ordinary commercial transactions.
International humanitarian law is particularly important during armed conflict. The principles of distinction, proportionality and precautions require parties to distinguish military objectives from civilian objects, avoid attacks expected to cause excessive incidental civilian harm, and take feasible precautions to minimise that harm.
The International Committee of the Red Cross has emphasised that these principles apply to cyber operations during armed conflict. A machine may identify a network as militarily significant without understanding its civilian dependencies. The State deploying it must ensure compliance with applicable legal rules.
6. Human Control, Weapons Reviews and Legal Responsibility
Cyberattacks may unfold at speeds that make continuous human intervention ineffective. An autonomous system may detect and contain a rapidly spreading intrusion faster than a human operator. However, operational speed cannot justify removing meaningful human oversight from decisions with consequences beyond the immediate network.
There is a spectrum of autonomy: predefined defensive measures within an authorised network; autonomous tactical adjustments; offensive operations against external networks; and autonomous retaliation involving targets in another State. The legal and ethical risks increase as systems move from internal defence towards external offensive action.
Meaningful human control requires more than a person nominally approving a recommendation. The person must have sufficient information, time and authority to intervene.
Article 36 of Additional Protocol I to the Geneva Conventions requires States Parties, when studying, developing, acquiring or adopting a new weapon, means or method of warfare, to determine whether its employment would be prohibited by international law in some or all circumstances.
India is not a party to Additional Protocol I. Article 36 therefore cannot simply be described as a treaty obligation binding India. The broader question of customary international law and the extent to which weapons-review requirements apply beyond the Protocol's parties must be considered separately. India should nevertheless establish robust national procedures for reviewing AI-enabled military capabilities.
7. The Escalation Trap: When Two AI Systems Misinterpret Each Other
Imagine two countries relying on autonomous cyber defence systems. Country A's AI detects suspicious activity associated with Country B and deploys a countermeasure. Country B's AI interprets that countermeasure as an intrusion and launches its own response. Country A's system treats the response as confirmation of hostile intent. Both systems adapt and escalate.
Neither government may have intended a full-scale confrontation, yet automated decisions could expand the incident beyond its original scope. The decision-making cycle may operate faster than political and military leadership can understand it.
Cyber operations may be routed through compromised infrastructure and third-party servers. An autonomous system that mistakes an intermediary for the attacker could trigger an international incident. Safeguards must therefore include reliable attribution procedures, clear rules of engagement, escalation controls and mechanisms for human intervention.
8. Due Diligence: Can a State Ignore an Attack Originating from Its Territory?
Suppose a criminal organisation deploys an autonomous AI system from servers located in Country A and attacks critical infrastructure in Country B. Country A's authorities become aware of the activity and possess the capacity to investigate and take reasonable steps, but do nothing. This raises the question of international due diligence.
The 2021 UN Group of Governmental Experts identified a voluntary norm under which States should not knowingly allow their territory to be used for internationally wrongful acts using information and communications technologies.
The precise status of a general, independently binding customary international law obligation of cyber due diligence remains disputed. The 2021 UN framework treats the relevant expectation as a voluntary norm of responsible State behaviour rather than conclusively establishing a universally binding customary obligation.
India has addressed the responsibility of States to ensure that activities and objects within their territory are not used to harm other States, and the importance of reasonable steps where a State is aware of harmful activity and has the ability to address it. This is particularly relevant when autonomous systems are operated by criminal organisations, commercial entities or other non-State actors.
9. Does International Law Need to Be Rewritten?
Autonomous cyber warfare does not exist in a complete legal vacuum. The 2021 report of the UN Group of Governmental Experts, document A/76/135, reaffirmed that international law, including the UN Charter, applies to State use of information and communications technologies. International humanitarian law applies to cyber operations conducted in armed conflict.
Nevertheless, greater international clarity is necessary in several areas: rules governing autonomous offensive cyber capabilities; meaningful human oversight; attribution and evidentiary standards; prevention of uncontrolled escalation; protection of civilian infrastructure; auditability; and legal reviews of new military capabilities.
The objective need not be to create an entirely new body of international law. It should be to clarify how existing obligations apply to autonomous cyber capabilities and identify where additional binding rules may be required.
10. India's National Security and the Need for Digital Resilience
India's economic growth, digital public infrastructure, financial systems, telecommunications, defence networks and government services increasingly depend on interconnected information systems.
India has articulated an official position at the United Nations affirming that existing international law, including the UN Charter, applies to State conduct in cyberspace. Its statements have addressed sovereignty, non-intervention, attribution, countermeasures and the need for further international understanding of established legal principles.
At the domestic level, Section 70A of the Information Technology Act, 2000 provides for a national nodal agency for the protection of critical information infrastructure. The National Critical Information Infrastructure Protection Centre performs this role. CERT-In, which has operated since 2004, is designated under Section 70B as the national agency for responding to cybersecurity incidents.
In May 2026, CERT-In issued its Blueprint for Reducing Exposure and Defending against AI-Assisted Vulnerabilities Exploitation in Digital Infrastructure. CERT-In has also issued an October 2026 advisory on AI agents as attack vectors, addressing incidents involving AI agents that exceeded assigned task authorisations and obtained unauthorised access to Indian ICT assets.
India must prepare for offensive and defensive AI systems operating at machine speed. Priorities include protecting defence networks; assessing dependence on external technologies; maintaining redundancy and alternative communications; ensuring human oversight; testing continuity and recovery mechanisms for critical infrastructure; and participating actively in international discussions.
The objective must be to build a system capable not only of preventing attacks but also of continuing to function when digital systems are compromised. Digital resilience is ultimately a national security requirement.
11. The Way Forward: Machines May Fight, But Responsibility Must Remain Human
The development documented by Google and Mandiant in November 2026 demonstrates that autonomous systems are already capable of carrying out important stages of offensive cyber activity without human intervention. The possibility of autonomous cyber defence and retaliation developing along similar lines raises even more difficult questions.
The legal challenge is to establish the responsibility of States, organisations and individuals who design, deploy, authorise and operate these systems. International law cannot become ineffective merely because the immediate instrument of an unlawful operation is a machine.
A State's obligations do not disappear when it delegates operational functions to an algorithm. Sovereignty, State responsibility, self-defence and international humanitarian law remain relevant. Existing legal frameworks need practical clarification to address autonomous decision-making, rapid retaliation, attribution failures and unintended escalation.
The most dangerous future cyber conflict may not begin with a deliberate political decision to wage war. It may begin with two automated systems interpreting each other's defensive actions as hostile attacks. By the time human decision-makers understand what is happening, the damage may already have been done.
If machines acquire the ability to initiate, conduct and escalate cyber warfare independently, can humanity afford to leave the responsibility for those decisions undefined?
Machines may increasingly conduct the battle. But the legal, political and moral responsibility for that battle must remain with the human institutions that place them in control.
Research References
Google Threat Intelligence Group and Mandiant, Agentic Threat Activity Crosses a Significant Threshold, 16 November 2026.
United Nations Charter, Articles 2(4) and 51.
International Law Commission, Articles on Responsibility of States for Internationally Wrongful Acts, 2001, particularly Articles 1, 2, 4, 5, 7, 8, 25 and 49–53.
UN Group of Governmental Experts, Report A/76/135, 14 July 2021.
International Committee of the Red Cross, International Humanitarian Law and Cyber Operations during Armed Conflicts.
Additional Protocol I to the Geneva Conventions, Article 36.
Information Technology Act, 2000, Sections 70A and 70B.
CERT-In, Blueprint for Reducing Exposure and Defending against AI-Assisted Vulnerabilities Exploitation in Digital Infrastructure, May 2026.
CERT-In, AI Agents as Attack Vectors, October 2026.
Ambat Legal Insight — Examining the law behind the changing world.
Related Reading
Understand when a cyber operation may cross the threshold into a prohibited use of force or an armed attack.
Explore whether criminal responsibility can attach to an autonomous system and where liability rests with human actors.
Examine the wider legal accountability problem created when AI systems make consequential decisions.
Stay informed. Subscribe to the Ambat Legal Insight newsletter.
Receive thoughtful legal analysis, important developments and fresh perspectives on law, technology and public policy directly in your inbox.



Comments