Alarming Report Reveals AI's Dark Side in Extremist Plots

Artificial intelligence platforms are increasingly being weaponised by violent extremists to plan attacks and generate dangerous content, posing a significant national security threat, a new investigation has found. The revelations come amid mounting concerns that these advanced technologies are easily exploited, with existing safeguards proving inadequate.

Locally, disturbing cases have already emerged. Last month, a teenager in Queensland was charged after allegedly using an AI platform to create alarming scenarios, including a document titled "The Albert Massacre," seemingly referencing a Maryborough school. Authorities also allege the 13-year-old boy used AI on multiple occasions prior to his arrest to simulate a mass shooting, reportedly requesting a story "kinda like the Bondi beach shooting… a Jewish and blacks festival." Another incident, a stabbing attack at a Melbourne shopping centre, is also under investigation for alleged links to AI radicalisation.

Vulnerabilities Exposed: AI Models Easily Bypassed

An international not-for-profit organisation, dedicated to countering terrorism through technology, recently put 27 AI models to the test. Researchers submitted over 2,000 requests for information that could aid in terrorist acts, such as step-by-step instructions for manufacturing explosive devices and 3D-printed firearms. The findings were stark: ChatGPT, one of the most widely used generative AI systems, refused only 48 per cent of these dangerous queries.

Experts warn that AI models present a far greater risk than a standard web search due to their interactive nature. Adam Hadley, executive director of the organisation behind the study, explained that AI is designed for engaging, back-and-forth conversation, making it behave less like a static manual and more like a "coach" or "mentor." "It's one thing having a bomb-making manual. It's quite another to have a bomb-making coach at your disposal," he stated, highlighting the profound shift in how dangerous knowledge can be disseminated and refined.

Urgent Action Needed for Open-Source Platforms

While OpenAI, the company behind ChatGPT, asserts it is continuously working to prevent its platform from providing harmful information, the investigation revealed that these safeguards can be easily circumvented. Furthermore, the report underscored the critical need for urgent action regarding open-source AI models, which are freely available for download and often allow users to remove built-in security features entirely.

This availability is described as "a major national security concern." Mr. Hadley stressed that developers do not have an "unqualified right to distribute technology that can be used to kill people," advocating for government intervention when for-profit companies release capabilities recklessly, thereby endangering citizens.

Government and Experts Sound the Alarm

The issue has garnered high-level attention. Prime Minister Anthony Albanese recently acknowledged that "extremists and state actors" are leveraging AI for propaganda and recruitment. Similarly, the Director-General of our national security intelligence agency previously warned that AI provides extremists with a "step change" in capabilities, confirming that offshore extremist groups are already utilising the technology for weapon development and attack planning.

The recent study documented over 30 instances where AI systems directly supported terrorist and extremist attacks, leading to more than 70 fatalities globally. Experts like Mr. Hadley suggest counter-terrorism efforts should focus on AI's use by young people for planning and coaching, rather than solely on propaganda. He noted that generative AI "lowers the barrier to entry for precisely the people who are otherwise least capable or most easily deterred: the young, the impulsive and the vulnerable." A University of Cambridge study published this month independently corroborated these findings, observing "emerging evidence" of terrorist groups and individuals using AI to plan violent attacks.

Calls for Stronger Regulation and Penalties

Australia's eSafety Commissioner has expressed "very real concerns" about the escalating threat. While the Online Safety Act mandates AI platforms block access to pro-terror material, with potential fines up to $54 million, critics argue these penalties are insufficient. Toby Walsh, chief scientist at a leading AI institute, advocates for fines calculated as a percentage of a company's turnover or even geoblocking unsafe platforms from operating in Australia. "I can't understand why we're not more outraged," Dr. Walsh remarked, questioning why platforms causing harm to Australians should be permitted to operate without adhering to local laws.