OpenAI admitted that new versions of ChatGPT will simplify the creation of biological weapons
OpenAI company management admitted, that the next generation of artificial intelligence, particularly the ChatGPT model, could create new risks, including the possibility of being used to develop biological weapons. This would potentially enable people without special scientific training to create dangerous substances.
Although the exact timing of the appearance of such high-risk models has not been announced, OpenAI’s head of security systems, Johannes Heideke, emphasized that some of o3’s successors will reach this level.
“We’re not yet in a world where there’s something new, completely unknown, that creates biological threats that didn’t exist before. We’re more concerned with recreating things that experts are already very familiar with.”, Heideke explained.
One of the key challenges is that the capabilities that allow large language models to achieve breakthroughs in medicine can also be used for malicious purposes.
Heideke acknowledged that OpenAI and other developers need systems that can detect and prevent malicious use of AI with high accuracy. The target level of reliability is more than 99%, ideally no more than one incident per hundred thousand. At the same time, human control plays a special role: monitoring teams must quickly detect and neutralize threats that bypass automatic protection mechanisms.
We will remind, in 2024, OpenAI announced the creation of an early warning system that blocks attempts to use ChatGPT to obtain data that could contribute to the creation of biological weapons. In April, the o3 and o4-mini monitoring system was implemented, which monitors potentially dangerous requests related to chemical or biological threats. Such measures are aimed at preventing the AI from providing instructions that could be used to develop real weapons.
