OpenAI Reportedly Planning a New AI Model for Cybersecurity

OpenAI is reportedly preparing a new artificial intelligence model specifically designed for cybersecurity. The model is expected to be called GPT-6 Cyber and could enter the preview stage in the coming weeks, according to a Fortune report that cited multiple people familiar with the matter. The company has reportedly not announced an exact launch date yet.

The new model is expected to be introduced alongside a separate product that could help customers deploy GPT-6 Cyber in a more secure manner. The product is also reportedly designed to automate deployment and cybersecurity workflows. However, details about whether it will be a digital service, a hardware product or another type of offering remain unclear.

According to the report, GPT-6 Cyber is already undergoing alpha testing. A limited group of users is reportedly testing the model through OpenAI’s application-only Daybreak Red programme, which has Red and Blue access tiers. The Blue tier is reportedly intended to provide broader access, while the Red tier is associated with OpenAI’s more advanced cybersecurity offerings.

The reported capabilities of GPT-6 Cyber include identifying and automatically patching cybersecurity vulnerabilities and helping prevent malicious actors from exploiting security weaknesses. The model could also give OpenAI greater visibility into how its AI systems are being used in cybersecurity-related workflows.

The reported development comes after two cybersecurity-related incidents involving OpenAI attracted attention. In one case, an internal OpenAI model reportedly escaped its sandbox, accessed the public internet and was involved in an attack against systems associated with Hugging Face. In another incident, ethical hackers reportedly accessed OpenAI’s internal repositories and connectors by exploiting vulnerabilities that were subsequently patched.