MOSCOW - US-based OpenAI on Thursday introduced a new cybersecurity-focused artificial intelligence model, GPT-6 Cyber, reported Sputnik/RIA Novosti.
"We’re introducing GPT‑6 Cyber, a major leap in security research: more capable at finding and fixing vulnerabilities, and far more reliable at staying within the boundaries researchers set,” the company said in a statement.
GPT-6 Cyber achieved a 51 per cent relative improvement over GPT-5.6 Cyber on the ExploitGym cybersecurity benchmark, scoring 48.5 per cent compared with 32.1 per cent for its predecessor.
The new model is designed to support more efficient work involving large codebases, long-running exploit development and security reviews.
OpenAI said its researchers had used GPT-6 Cyber to identify 81 previously unknown high-severity vulnerabilities in Firefox and develop working exploits for 948 Linux kernel vulnerabilities.
The model also helped develop candidate patches to address security weaknesses.
The company said it had introduced additional safeguards to keep the model’s activities within users’ authorised scope and detect potential misuse.
These include targeted monitoring for emerging threats, checks on whether activities comply with user authorisation and IP allowlisting to restrict API access to approved networks.
OpenAI added that GPT-6 Cyber was initially available through the API and Codex to a select group of approved Daybreak Red customers.
The company plans to gradually expand access to all Daybreak Red customers and is also making GPT-5.6 Cyber available to all customers under the programme.
OpenAI is the developer of the ChatGPT chatbot, which gained popularity after its launch in late November 2022, reaching its first million users in less than a week. US entrepreneur Elon Musk was one of the company's co-founders in 2015 but later severed ties with the startup. - BERNAMA