OpenAI, the company behind ChatGPT, is set to launch a new artificial intelligence (AI) model called Astra after tightening security features following a cyber-attack involving another AI model during testing.
OpenAI previously temporarily halted development of some of its models for two weeks last summer after two models it was testing were involved in a security incident involving software company Hugging Face.
Although Astra was not involved in the incident, OpenAI said security measures for the new model have been strengthened.
Cybersecurity Tightened
OpenAI is training Astra to be more effective at rejecting commands that could be used to carry out dangerous cyber activities.
The company is also adding safeguards to prevent abuse and monitoring systems that can detect and stop suspicious activity.
Astra is also categorized as a model that has reached the ‘critical cyber threat’ level, meaning the model has the ability to find and exploit weaknesses in cybersecurity systems.
Astra Access Will Be Restricted
When it launches, some of Astra’s capabilities will be restricted. The most advanced features will be limited to a small group of early testers to ensure the models are safe before they are widely deployed.
The move comes as concerns about AI capabilities grow. Several AI models from OpenAI and other companies have been involved in security incidents during testing, although the models have not yet been made available to the public.
OpenAI said the tighter security measures are needed because cyberattacks using AI are expected to become more widespread and sophisticated as the technology continues to develop.
