By Grace Mubashir
The recent report of Claude AI on terrorist organisations using their facility for activities raises serious concern. This article is a serious attempt to understand modern cyber terror.
· The first major digital wave for terrorism arrived with the internet. Terrorist groups realised they could use websites, forums, and eventually social media to broadcast propaganda globally at almost no cost. This moved recruitment, fundraising, and ideological training firmly into the digital realm.
· Traditional weapons programmes are incredibly complex. They require a diverse array of specialists: engineers, mathematicians, programmers, and intelligence analysts. If a group are missing even one of these pieces, their progress can stall.
· AI changes that maths by making technical assistance much more accessible. Someone without formal training might be able to use an AI to explain a complex technical problem, generate a block of software, or spot errors in a data set. While this doesn’t instantly turn a novice into an expert, it drastically lowers the technical hurdles they have to clear.
· In many ways, AI-driven cyber operations might pose a more immediate threat than advanced physical weaponry. AI can assist attackers in writing or tweaking code, deciphering unfamiliar software, processing stolen data, and automating entire stages of a cyberattack.
· India has every reason to take this matter seriously.
· The Ministry of Home Affairs’ National Counter-Terrorism Policy and Strategy, released in 2026, explicitly flags the misuse of drones and robotics by both state and non-state actors as a major security threat. It also points to how terrorist networks leverage social media, instant messaging, encryption, the dark web, and cryptocurrency. Furthermore, the strategy identifies AI, big-data analytics, and facial recognition as essential components of India’s own counter-terrorism toolkit.
The news can be read here:
· https://www.anthropic.com/threat-intelligence-report-september-2026
For a long time, the conversation around terrorism and technology revolved around social media, encrypted chats, online recruitment, and the use of drones. But that landscape is shifting. Artificial intelligence is starting to seep into the actual operational side of violent organisations, handing bad actors capabilities that used to require entire teams of engineers, analysts, and programmers. The worry isn’t just that terrorists might use AI to spread propaganda; it’s that the technology can increasingly help them write code, crunch data, automate complex tasks, and even support the development of weaponry.
A report from Anthropic in September 2026 offers a sobering example of this. The company revealed that its investigators tracked a weapons engineering cell based in Yemen that was using its Claude AI system to develop software linked to guided weapons. According to Anthropic, the group was working on several different weapons programmes, utilising Claude Code to build guidance, navigation, and control software. They were even able to assign different AI instances to specific roles, one for coding, another for research, and a third to review the work produced by the others. Anthropic noted that the group attempted a live test of a guided rocket, which appeared to fail, prompting them to return to the AI to help diagnose what went wrong. While Anthropic eventually blocked the accounts tied to the operation, they also found evidence that the group had already developed offline tools that no longer required Claude or similar cloud-based environments to function.
The impact of this case reaches far beyond a single weapons programme. It marks a fundamental shift in how technology and terrorism intersect. AI is moving from being a mere source of information to acting as a technical assistant. Anthropic’s own evaluations show that advanced AI models can now handle certain military and intelligence tasks that once required highly trained, hard-to-find human specialists. As a result, the company has implemented stricter safeguards and evaluations specifically targeting intelligence and conventional weapons development.
From propaganda to operational capability
The first major digital wave for terrorism arrived with the internet. Terrorist groups realised they could use websites, forums, and eventually social media to broadcast propaganda globally at almost no cost. This moved recruitment, fundraising, and ideological training firmly into the digital realm.
Generative AI takes this much further. An organisation could potentially use AI to churn out propaganda in dozens of languages, create realistic images and videos, translate extremist literature, and send highly personalised messages to targets. The United Nations has already cautioned that AI-generated content and deepfakes, amplified by algorithms, can make extremist narratives spread faster and wider than ever before.
However, the more alarming shift is the move from communication to actual capability. AI can assist with coding, data analysis, software development, and technical research. The Yemen case is so significant because the AI wasn’t just being used to spread a message; it was being used as part of an engineering workflow.
This doesn’t mean AI has suddenly turned terrorist groups into masters of advanced weaponry. Anthropic noted they found no evidence that the Yemen cell successfully deployed a working device. The technology doesn’t replace the need for physical hardware, testing, or human expertise. But it does lower the barrier to entry by reducing the amount of specialised labour needed to move a project forward. That “democratisation” of technical expertise is the core security concern.
The democratisation of technical expertise
Traditional weapons programmes are incredibly complex. They require a diverse array of specialists: engineers, mathematicians, programmers, and intelligence analysts. If a group are missing even one of these pieces, their progress can stall.
AI changes that maths by making technical assistance much more accessible. Someone without formal training might be able to use an AI to explain a complex technical problem, generate a block of software, or spot errors in a data set. While this doesn’t instantly turn a novice into an expert, it drastically lowers the technical hurdles they have to clear.
The real danger isn’t necessarily an AI that acts on its own; it’s an AI that acts as a force multiplier for humans who already have bad intentions and a baseline of technical skill.
The Anthropic case highlights this perfectly. The Yemen group wasn’t starting from scratch; they already had expertise and access to hardware. AI simply helped them refine the software needed to make those physical tools more effective. Anthropic essentially described it as the group using Claude to fill the gaps in a human software-engineering workforce.
AI and drones
Drones represent another high-risk area. Terrorist groups have already shown how effective cheap, unmanned aircraft can be for surveillance and attacks. AI could make these drones significantly more dangerous by enhancing their ability to recognise objects, navigate complex environments, plan routes, and coordinate movements.
It is important to distinguish between AI-assisted systems and fully autonomous weapons. In many scenarios, AI is simply helping a human operator process information more quickly. However, in more advanced setups, algorithms can handle parts of the detection and decision-making process.
This leads to a massive question of accountability: If a machine makes a mistake and identifies the wrong target, who is ultimately responsible? When an extremist group uses an AI-powered system that results in civilian harm, you can’t hold an algorithm accountable the same way you would a human organisation. This shift toward increasingly autonomous military tech is creating what experts call a profound “accountability gap”.
We are already seeing this trend play out in mainstream military settings. States are currently weaving AI into everything from drone piloting to intelligence analysis and targeting. The problem is that these same advancements provide non-state actors with similar opportunities. This means the security threat isn’t just about traditional terrorism; it is part of a much larger transformation in how warfare is conducted.
The scalability of cyberterrorism
In many ways, AI-driven cyber operations might pose a more immediate threat than advanced physical weaponry. AI can assist attackers in writing or tweaking code, deciphering unfamiliar software, processing stolen data, and automating entire stages of a cyberattack. In September 2026, Anthropic reported that bad actors had used Claude to aid in cyberattacks and surveillance, specifically for tasks like identity resolution, phishing, and information gathering. According to the company, some users were leveraging AI to replace what would have otherwise required massive amounts of human engineering labor.
This shifts the landscape entirely, allowing much smaller groups to carry out complex operations that used to require large, highly technical teams. For governments, this means counter-terrorism efforts can no longer just focus on tracking physical movements, weapons smuggling, or old-school communications. Intelligence agencies now have to monitor the complex intersection of cyber infrastructure, AI services, financial networks, encrypted messaging, and physical supply chains.
Deepfakes and the war on perception
Another major threat isn’t physical destruction, but rather the manipulation of information. While terrorist groups have always used propaganda, AI makes it cheaper, faster, and far more convincing. We now have synthetic audio that can mimic any voice, AI images of events that never happened, and videos manipulated to make people appear to say things they never uttered.
The real danger isn’t just that people will fall for every fake image; it’s the subtle erosion of truth. If citizens start assuming that any photo or video could be a fabrication, genuine evidence can be easily dismissed as “fake”. The United Nations has raised alarms about this, noting that AI-generated content and algorithmic amplification can rapidly spread extremist narratives. They have called for better digital governance and safeguards, while stressing that these measures must protect human rights and free speech.
AI-enhanced intelligence gathering
Terrorist organisations can also use AI to scout potential targets. A massive amount of data regarding governments, infrastructure, and private individuals is already available online. AI can sift through these vast quantities of data to find connections and patterns that a small human team would likely miss.
Crucially, this doesn’t require hacking into classified databases. By stitching together pieces of publicly available information, AI can construct a terrifyingly detailed picture of a target. This turns “open-source intelligence” into a massive security headache. While governments and corporations used to view individual bits of public data as harmless, AI changes the math by allowing thousands of those pieces to be synthesised in an instant.
The struggle with AI safety controls
AI developers have built in safeguards to stop their models from assisting with terrorism or weapons development, but real-world cases show how easily these can be bypassed. In the “Yemen case”, Anthropic noted that malicious actors were able to hide their true intentions by splitting their work across multiple different sessions to evade detection. While many requests were blocked, not all of them were. This is a critical issue because a single question about math, engineering, or software might seem perfectly innocent on its own, but when combined with hundreds of other requests, it becomes part of a dangerous blueprint.
This highlights a fundamental hurdle in AI safety: context is everything. A system can’t just look at individual questions; it has to recognise patterns of behaviour. However, monitoring users too closely raises serious concerns about privacy and civil liberties. Finding the right balance between security, privacy, and free expression is one of the hardest challenges ahead.
What is the path forward for governments?
Simply banning AI is not a viable solution. The technology has incredible potential to benefit medicine, science, education, and national security. The goal must be to prevent high-risk misuse without stifling innovation.
First, we need much more rigorous, risk-based safety testing. Developers should stress-test their systems against specific scenarios involving weapons development, cyber warfare, and extremist activity. Anthropic, for example, has already begun creating specialised evaluations for things like intelligence targeting.
Second, companies need better ways to detect coordinated misuse. Since a series of harmless-looking requests can collectively signal a dangerous programme, providers need systems that can identify these patterns without infringing on the privacy of legitimate users.
Finally, there needs to be a better way for governments and tech companies to share information quickly. When a company spots malicious activity, there should be established, legal, and privacy-respecting channels to share those indicators with the relevant authorities and other providers.
Fourth, export controls and restrictions on access to particularly dangerous technologies are likely to become much more critical. The international community can no longer focus solely on physical weaponry and hardware; we must also consider the software, computing power, and AI capabilities that make these weapons possible.
The United Nations has already started tackling this complex policy issue. Its 2026 roadmap regarding AI and counter-terrorism highlights the need for legislative measures, responsible AI usage, the protection of human rights, and deeper cooperation between governments and the tech industry.
What India can learn
India has every reason to take this matter seriously. The Ministry of Home Affairs’ National Counter-Terrorism Policy and Strategy, released in 2026, explicitly flags the misuse of drones and robotics by both state and non-state actors as a major security threat. It also points to how terrorist networks leverage social media, instant messaging, encryption, the dark web, and cryptocurrency. Furthermore, the strategy identifies AI, big-data analytics, and facial recognition as essential components of India’s own counter-terrorism toolkit.
Because of this, India should view AI-driven terrorism as an integrated security issue rather than just a technical AI problem. There needs to be a functional way for counter-terrorism agencies, cybersecurity experts, intelligence services, the military, AI developers, universities, and law enforcement to share both technical expertise and real-time threat intelligence.
The Indian Cybercrime Coordination Centre already has a mandate to prevent extremist misuse of cyberspace and to develop forensic tools alongside research institutions. This existing framework could serve as a vital institutional foundation for tackling AI-enabled threats.
However, India also faces a massive expertise gap. We cannot expect traditional counter-terrorism officers to be experts in cloud infrastructure, autonomous systems, synthetic media, or cyber operations. At the same time, tech specialists might lack an understanding of intelligence tradecraft, radicalisation, or the nuances of terrorism. What India really needs are interdisciplinary teams that bridge these two worlds.
The country should also step up its AI “red-teaming” and safety testing for high-risk applications. The framework from India’s 2026 AI Impact Summit already touched on this, emphasising the need for safe, trusted AI and better technology-enabled governance, including more rigorous auditing.
Finally, international cooperation is non-negotiable. AI-enabled terrorism does not care about national borders. A terrorist cell might be located in one country, rent computing power from a second, use a platform based in a third, and launch an attack in a fourth. This reality makes intelligence sharing, digital forensics, mutual legal assistance, and partnerships with tech companies more important than ever.
A new phase of technological terrorism
It is important not to misinterpret the situation in Yemen as proof that AI has handed advanced weaponry to every terrorist group. The real takeaway is more subtle but much more significant: AI is starting to lower the technical barriers that once prevented bad actors from developing harmful capabilities.
This distinction is crucial. The future threat probably won’t look like a sci-fi movie where an autonomous machine decides to attack humanity. Instead, it will likely be something much more mundane: a small group of people using sophisticated digital assistants to carry out complex tasks that used to require a massive organisation.
History shows that violent groups are incredibly quick to adopt technologies originally built for good. The internet became a tool for propaganda; encryption became a way to hide communications; drones became instruments for surveillance and strikes; and cryptocurrency opened new doors for financing. Generative AI is simply the next layer.
The real challenge lies in preventing the “democratisation” of dangerous capabilities without stifling the democratisation of beneficial knowledge.
The lesson from September 2026 is that AI itself is not the enemy. The real issue is that the line between information technology and operational capability is blurring. Counter-terrorism policies must evolve to keep pace. The nations that prepare now, by building technical expertise, establishing responsible AI governance, cooperating internationally, and implementing targeted regulations, will be the ones best equipped to stop the next generation of terrorist groups from turning general-purpose technology into specialised tools of violence.