Forget the model. When it comes to cybersecurity, it’s all about the harness - CyberScoop

A new 'harness' tool designed to pair with OpenAI's ChatGPT 5.5 and GPT 5.5-Cyber models has been demonstrated for autonomous hacking tests. The tool showed accelerated reasoning and attack capabilities, according to a report published by CyberScoop.

The harness integrates the two OpenAI models to autonomously identify and exploit vulnerabilities. The GPT 5.5-Cyber variant is fine-tuned for cybersecurity applications. The demonstration indicated that the harness enables the models to reason through attack strategies more quickly than previous methods. The report did not specify the exact performance metrics but described the results as showing 'accelerated reasoning and attack capabilities.'

The findings come amid growing interest in using large language models for cybersecurity both offensively and defensively. The report did not name the researchers or institutions involved, nor did it specify whether the harness is publicly available.