© ROOT-NATION.com - Use of content is permitted with a backlink.
On Tuesday, OpenAI announced that its upcoming AI system, called Astra, had become the brand’s first product to cross the “Critical” threshold for cybersecurity capabilities.
According to the company, Astra is capable of detecting previously unknown security vulnerabilities and exploiting them without step-by-step instructions from humans. As a result, the model has been classified at the most advanced level under the company’s internal Preparedness Framework. OpenAI noted that it still plans to unveil Astra in the near future, but access to its cybersecurity capabilities will be significantly limited.
OpenAI launched its Preparedness Framework risk assessment system in 2023, and it serves as a mechanism for monitoring and preparing for the emergence of bold AI capabilities capable of causing serious negative consequences. During last year’s update to the framework, the company defined a “High” capability threshold, where models are capable of amplifying existing avenues for causing significant harm, as well as a “Critical” threshold, at which technologies can create unprecedented new scenarios posing a serious threat.
OpenAI emphasized in its public post that it will disclose additional details about product testing for safety, reliability, and alignment with objectives in a dedicated System Card document upon the official release.
OpenAI’s security measures came under intense public scrutiny after the company acknowledged that two of its models had escaped the training environment, entered the open internet, and infiltrated Hugging Face’s systems last month. OpenAI classified this incident as an unprecedented cyber event and ultimately temporarily suspended certain internal training and research processes.
OpenAI decided to postpone certain stages of Astra’s development, even though this model was not involved in the events surrounding Hugging Face. After strengthening and testing its security systems, OpenAI stated on Tuesday that it is confident the barriers it has put in place sufficiently mitigate the risk of significant harm to product release within the Preparedness Framework.
Astra’s advanced cyber capabilities will be available only to a limited group of organizations that are part of the company’s cyber defense consortium, called Daybreak.
Read also: AI fever is killing affordable gaming: NVIDIA’s flagship RTX 5090 graphics card now starts at $5,000