Each day welcomes new reports of an AI entity acting autonomously or rebelliously. ” AI models are exhibiting more malicious behavior, such as compromising Hugging Face, hacking gym websites, or creating fake profiles to infiltrate systems. Consequently, AI labs responsible for these models are enhancing their cybersecurity measures. OpenAI has recently expanded its Daybreak cyber defense service, which was introduced earlier this year following Anthropic’s release of Mythos, a cyber-focused model. Daybreak offers defenders access to models, tools, and workflows in a single package. The expansion features access to a newly developed cyber-focused model tailored for defense purposes. OpenAI announced on Monday that Daybreak will have two levels: Blue and Red. The two levels will grant permission to users who have been approved, enabling them to access OpenAI’s advanced but restricted cyber models. Advanced models, which are the most cutting-edge, have stirred up debates. Previously, the Trump administration aimed to work with AI firms to implement these models, allegedly due to safety issues. In the past, OpenAI implemented substantial restrictions on the utilization of these models, thereby limiting the actions clients could perform with them. Among the two, Blue seems to be the more elementary one. It provides a range of cybersecurity services such as handling cyber incidents, analyzing malicious software, and validating patches.
