New OpenAI AI models can be used for chemical weapons
New models of artificial intelligence developed by the OpenAI company, announced on September 12, can be used to create biological weapons, reports the Financial Times. The company’s system document notes that the risk of creating chemical, biological, radiological, and nuclear weapons has increased to the “medium” level, which is the highest level of risk assigned to OpenAI models today.
The new class of AI, known as “o1”, has improved abilities to solve complex mathematical problems and answer scientific questions. OpenAI tested “o1” on the qualification exam for the International Mathematical Olympiad. At that time, the previous model solved only 13% of problems correctly, “o1” scored 83%.
According to the creators, the “o1” models are capable of performing step-by-step logical thinking. Experts warn that this increases the risk of their use by criminals.
California Governor Gavin Newsom should soon consider a bill that would require AI developers to take measures to minimize the risks of using these technologies to create biological weapons.
Representatives of technology companies, including OpenAI, express fears that the adoption of the document will slow down the development of AI.
Technical director of OpenAI Mira Murata told the Financial Times that the “o1” model showed significantly better security results compared to previous versions of the AI.
The preview version of the “o1” model is already available to owners of a paid ChatGPT subscription.
More news from Radio Liberty:
