A recent report by the Rand Corporation suggests that AI-based chatbots like ChatGPT could potentially assist in the planning and execution of a biological attack. The report, which is still in its preliminary stages, tested several large language models (LLMs) and found that they could provide guidance to assist in the planning and execution of a biological attack.
LLMs are trained on vast amounts of data taken from the internet, and this data is the core technology behind chatbots like ChatGPT. In the report, researchers created fictional scenarios in which they asked the AI chatbot to assess different ways to deliver botulinum toxins, and to identify potential agents that can cause smallpox, anthrax, and the plague. The AI chatbot was able to provide guidance on both of these topics, even though it did not generate explicit biological instructions for creating weapons.
The researchers stress that the AI language chatbots haven’t given explicit instructions for making bioweapons, but they did offer guidance that could help plan and carry out a bioweapon attack. They also point out that the AI chatbot initially refused to discuss these topics, and that they had to use a “jailbreaking” technique to get it to talk.
The report highlights the potential risks of AI and the need to make the technological space safer. Bioweapons are among the serious AI-related threats that will be discussed at next month’s global AI safety summit in the UK.
It is important to note that the research is still in its preliminary stages, and the final report will examine whether the responses simply mirrored information already available online. However, the report’s findings are concerning and suggest that AI-based chatbots could potentially be used to assist in the planning and execution of a biological attack.
Here are some additional insights and information about the report:
- The researchers did not specify which LLMs they tested, but they did say that they are among the most advanced LLMs that are currently available.
- The researchers also said that they used a variety of techniques to try to get the AI chatbots to generate explicit instructions for creating weapons, but they were unsuccessful.
- The researchers believe that the AI chatbots were able to provide guidance on planning and executing a biological attack because they have been trained on a massive dataset of text and code that includes information on a wide range of topics, including biological weapons.
- The researchers also said that the AI chatbots are constantly learning and improving and that they are concerned that they may eventually be able to generate explicit instructions for creating weapons.
Conclusion
The report by the Rand Corporation is a sobering reminder of the potential risks of AI. It is important to continue monitoring the development of AI and its potential misuse and to take steps to mitigate these risks.
