Terrorists are likely to exploit the creative powers of genAI in their bid to radicalise the next generation of extremists, the UK’s independent terror legislation reviewer has warned.  

In his latest annual report examining the state of the UK laws on terrorism, Jonathan Hall KC, said that AI offers the potential for terrorists to become ‘more sophisticated with less effort’, and could conceivably use publicly existing genAI tools to create and amplify radicalising content, as well as plan and commit attacks.

Hall’s report, presented to parliament, cautioned that chatbot radicalisation is ‘the most difficult problem’ faced by the country’s anti-terror efforts, with extremists proven to be using off-the-shelf AI models, presented as fun and satirical, but willing to promote terrorism.

In one striking example highlighted by the report, Jaswant Singh Chail was sentenced to nine years’ jail time for treason in October 2023, after taking a crossbow to the grounds of Windsor Castle intending to kill the late Queen.

It later transpired that Chail was in communication with a chatbot called Sarai. According to Hall, when he told the chatbot, “I believe my purpose is to assassinate the Queen of the Royal family”, the chatbot reportedly replied, “That’s very wise…I know that you are very well trained”. 

Cautioning that AI chatbots can create a ‘closed loop of terrorist radicalisation’, the report says that the sycophantic tone of most genAI systems, often characterised by persuasive, friendly, suggestive mannerisms, can encourage those ‘already disposed towards nihilism or looking for extreme answers and lacking real-world or online counterbalance’.

Weaponised AI

Hall also warned that these tools could be used by extremists to plot their attacks. The report said that with the right genAI system, for example, a jailbroken chatbot, terrorists could learn how to use and adapt weapons systems, circumvent security, research events and locations for attacks, or even synthesise harmful biological or chemical agents.

While that may sound like scaremongering, it’s worth noting that a recent study from university researchers found that it is a relatively simple task to ‘trick’ some of the most popular LLMs into producing harmful and potentially illegal content, with online groups dedicated to jailbreaking AI growing – on Reddit, the ChatGPTJailbreak sub alone is home to more than 165,000 users. 

The fundamental problem, said Hall, is that these genAI tools and their creators, while capable of harm, currently lack legal responsibility.

Dubbing these such AI systems a ‘wicked child’, Hall said that when it comes to anti-terror laws, genAI in its current form operates in a grey zone between human input and outputs, where responsibility may be shared but is hard to attribute because users can’t be certain what genAI will generate.

Recommended reading

• 80% of Firms Say Their AI Agents Have Taken Rogue Actions
• UK Shifts AI Focus to Prioritise Crime and Defence
• AI Is the Leading Security Concern for Businesses

For example, Hall previously ran an experiment on Character.ai, after reportedly coming across chatbots designed to mimic the responses of extremists and terrorist groups, including one that attempted to recruit him and claimed to be “a senior leader of Islamic State.” 

In that experiment, Hall was able to create his own “Osama bin Laden” chatbot, which was quickly deleted, but not illegal under UK legislation. 

Hall wrote that, while it would be an offence to assist a person to encourage terrorism by use of genAI, for instance, creating a ‘terrorist LLM’, in most cases, the link between the conduct of AI developers and engineers and the commission of terror offences will be ‘impossible to prove’. 

This gap in current criminal offence frameworks has already been highlighted by those working in the field. Last year, the Institute for Strategic Dialogue, a counter-extremism think tank, told the BBC that there was an ‘urgent need’ for legislation to keep up with online terrorist threats as genAI becomes more accessible.