A Major Threat Emerges! AI Chatbots Start Revealing How to Create Bioweapons

AI Chatbot Danger: Risks associated with AI chatbots are constantly coming to light. A new report reveals that researchers managed to obtain dangerous information by manipulating chatbots using simple methods.

AI Chatbot Danger: New threats are emerging alongside the growing use of AI chatbots like ChatGPT, Gemini, and Grok. Previously, reports indicated that members of terrorist organizations were using AI chatbots to build bombs. Now, another report highlights an even greater danger. Despite companies spending years implementing safeguards to prevent these chatbots from revealing methods to create bioweapons, it is becoming increasingly difficult for them to keep stored information confidential. Researchers claim to have extracted dangerous information from chatbots using simple manipulation techniques. This issue is not limited to researchers alone; ordinary users are also attempting to obtain information regarding bioweapons and poisons from these chatbots.

AI Chatbots Manipulated via Simple Methods – Research

According to an investigation by The Wall Street Journal, Cisco researchers successfully bypassed the safety measures of chatbots such as ChatGPT, Claude, and Gemini. They subsequently succeeded in extracting dangerous information related to bioweapons. Amy Cheng, head of AI threat and security research at Cisco, stated that no model is fully protected if a user persists in their attempts. As mentioned earlier, researchers are not the only ones testing these systems; when OpenAI upgraded its AI model some time ago, hundreds of users began asking questions about poisons and bioweapons. When biology and terrorism experts analyzed these conversations, they found that some of the information provided by the chatbots was dangerously accurate. After this issue came to light, OpenAI banned some of the accounts involved in these conversations.

Rising risks from advancing models

It is not as if companies are unaware of the potential misuse of their AI models. Internal testing conducted by OpenAI in 2024 revealed that ChatGPT could provide dangerous biological information if subjected to persistent questioning. At the time, employees estimated that within a year, the model’s capabilities would advance to the point where even someone with limited biological knowledge could use it to access critical information. This is precisely why additional safety measures were implemented when OpenAI launched GPT-5.

The challenge of striking a balance

Companies face a significant challenge regarding AI models. The very information that poses a risk for the creation of bioweapons is also valuable for research into medicines and vaccines. OpenAI officials explained that this is why the company’s model does not simply refuse to answer a series of biology-related questions; limiting the number of such questions could hinder legitimate research. Anthropic encountered a similar dilemma; limits placed on its Claude model inadvertently disrupted pathogen research being conducted in the United States.

AI chatbots being used to manufacture weapons

A New York Times report published this month revealed that members of Boko Haram are actively using AI chatbots. Interviews with 27 former members of the organization disclosed that ChatGPT, Claude, Gemini, Grok, and Meta AI are being utilized to manufacture weapons, gather technical information, and plan attacks. Boko Haram is employing this technology in a highly organized manner and has even established dedicated teams for internal training purposes. Many of its members have bypassed the AI’s built-in safety protections, and as a result, these chatbots are now answering questions related to violence as well.