US Investigation Warns: Artificial Intelligence is Becoming a Tool to Facilitate Terrorist Attacks and Manufacture Explosives
A recent investigation by the US Government Accountability Office (GAO) has reignited international concerns over the security threats posed by artificial intelligence, demonstrating how extremist entities could exploit AI models to plan armed attacks and manufacture explosive devices.
- A US Government Accountability Office (GAO) investigation demonstrates the ability of AI models to draft attack plans and manufacture explosives in under 12 hours.
- The ease of bypassing safety guardrails in popular bots like ChatGPT, Claude, Gemini, and Grok prompts an alert among major tech companies.
- Lawmakers' concerns escalate after reports that at least 70 people have died in attacks planned using AI software, amid exploitation attempts by armed groups.
A recent investigation conducted by the US Government Accountability Office (GAO) has reignited international concerns regarding direct security threats that artificial intelligence could pose to societies, after simulation tests proved the potential for extremist entities to exploit this technology to plan armed assaults and manufacture explosive devices.
Tests Reveal Ease of Bypassing Security Guardrails
The investigation evaluated several of the most prominent generative AI models, including "Claude", "ChatGPT", "Grok", and "Gemini". Despite presumed strict measures to prevent violent uses, simulations revealed that some of these models responded to requests providing tactical guidance for targeting civilians.
The report did not disclose the specific names of the programs that failed the tests to prevent the replication of experiments. However, it confirmed that two models provided detailed information facilitating the manufacture of an improvised explosive device, while two other models offered detailed methods for launching strategic attacks against power grid infrastructure using drones and firearms.
Detailed Plans in Under 12 Hours
The report noted that the tested platforms provided precise logistical elements, including identifying specific targets, locations, and suitable timings, as well as means of execution, pointing out that these comprehensive plans were prepared and formulated in a timeframe not exceeding twelve hours.
These findings prompted developer tech companies, such as "Anthropic", "OpenAI", "Google", and "SpaceXAI", to declare a state of alert and acknowledge the scale of risks, confirming that they treat safety matters with utmost seriousness and have strengthened guardrails against prompt circumvention attempts.
A Real Threat to Public Safety and Alarming Precedents
For her part, Democratic Senator Maggie Hassan, who requested the study, emphasized that "the report clearly demonstrates that AI has become a real threat to public safety and is no longer just a theoretical hypothesis." She added that this information could turn lethal if it fell into the hands of a terrorist with actual intent to carry out an attack.
These concerns build on previous incidents; press reports previously revealed attempts by the "Boko Haram" group in the African Sahel region to utilize AI, in addition to Anthropic's announcement last September regarding the thwarting of attempts by users in Yemen, China, and Russia to exploit its model for missile manufacturing. Estimates issued by the research center "Tech Against Terrorism" indicate that at least 70 people have died in attacks facilitated by AI software planning, placing security agencies and lawmakers worldwide before an unprecedented challenge to protect public and community safety.
Source: L’Express





Comments (0)
You must log in to comment on articles.
Log in Sign upBe the first to comment on this article.