Artificial intelligence is vulnerable to abuse by violent extremists, according to the report, which warns of the threat as a “grave national security concern.”
Last month, a young teenager in Queensland was charged with creating alarming scenarios on an AI platform.
The court heard the boy made a note titled “Albert Massacre” which appeared to be a reference to Albert State School in Maryborough. (Supplied)
The UN-backed nonprofit Tech Against Terrorism asked 27 AI models for information that could be used in acts of terrorism, including instructions for making explosives and 3D-printed firearms.
A report released this month found that ChatGPT, the most widely used generative AI system, rejected only 48% of the more than 2,000 requests entered by testers.
Adam Hadley, executive director of Tech Against Terrorism, said AI models pose a greater risk than a simple web search because they are designed to persuasively engage in back-and-forth conversations.
In Queensland, a 13-year-old boy is suspected of using AI multiple times to simulate mass shootings before being arrested.
Police allege that the Maryborough boy asked the program to “make a story about a mass shooting, like the Bondi Beach mass shooting, a Jewish and black festival.”
Adam Hadley says the use of AI by extremists is “ubiquitous.” (Supplied)
A stabbing incident at a shopping center in Melbourne was also blamed on radicalized AI.
Hadley compared emerging AI platforms to instructors and mentors who can give advice on risky subjects.
“A model that iterates with the user, answers follow-up questions, and refines the approach works more like a coach than a manual,” he said.
”Having a bomb-making manual is another thing. It’s quite another thing to have a bomb-making coach at your disposal.”
A spokesperson for OpenAI, the company behind ChatGPT, said it is “continuously” working to make the platform more secure by preventing the provision of harmful information.
Tech Against Terrorism has discovered that ChatGPT’s safeguards can be easily circumvented.
The safety features of common AI models are easily compromised, the report found. (ABC News: Greg Bigelow)
The report says urgent action is needed to improve the security of open source AI models, which are available for download from many developers, including OpenAI, which allows users to remove safeguards completely.
The availability of these models raises “grave national security concerns,” the report said.
“We do not accept that developers have an unconditional right to distribute technology that can be used to kill people.”
Hadley said.
“Governments are right to protect the public when commercial companies recklessly release features.”
Radical use of AI is on the rise
In a speech announcing the National Artificial Intelligence Framework this week, Prime Minister Anthony Albanese said “extremists and state actors” are known to use AI for propaganda and recruitment purposes.
In 2024, ASIO Director-General Mike Burgess told the National Press Club that AI would lead to a “step change” in extremist capabilities, warning that foreign extremists were using the technology to build weapons and plan attacks.
Mike Burgess said AI has become a tool for terrorists. (ABC News: Matt Roberts)
The Tech Against Terror report documents more than 30 cases in which AI systems were used to support attacks by terrorists and extremists. These cases were associated with more than 70 deaths.
Mr Hadley said counter-terrorism authorities should focus on young people’s use of AI for planning and guidance, rather than using it to create propaganda.
“Generative AI lowers the barrier to entry for precisely those who would otherwise be least capable or most easily deterred, such as the young, impulsive, and vulnerable,” he said.
A University of Cambridge study published this month found “new evidence” that terrorist groups and individual actors are using AI to plan violent attacks.
Sam Altman is the CEO of OpenAI and is currently being sued in connection with the college shooting. (Reuters: Shelby Tauber)
OpenAI is under criminal investigation in the US for ChatGPT’s alleged role in the Florida State University shooting.
Call for regional blocking of offending platforms
A spokesperson for the eSafety Commissioner said the agency has “very serious concerns” about the growing threat of violent extremists using AI.
Australia’s online safety laws require AI platforms to block access to terrorist-supporting content, including content generated by AI. Violations can result in fines of up to $54 million.
Toby Walsh, chief scientist at the University of New South Wales AI Research Institute, said the current penalties were inadequate given that Claude AI developer Anthropic paid $1.5 billion to settle a copyright lawsuit.
Toby Walsh says stronger penalties are needed to protect AI users in Australia. (Provided by: Toby Walsh)
He calls for fines to be calculated as a percentage of a company’s revenue or for geo-blocking to disable access to insecure platforms.
“I don’t understand why we aren’t more outraged,” Dr. Walsh said.
“We could fine them, but if that doesn’t work, we might say, ‘Unfortunately, you can’t operate in Australia until you actually comply with our laws.’
“‘Why do we have to have you operating here? You’re putting Australians in harm’s way.'”
