The UK AI Security Institute has identified that OpenAI's upcoming GPT-5.5 model possesses cyberattack capabilities comparable to Anthropic's Claude, according to a report from decrypt.co.
Testing conducted by the institute suggests that the advanced large language model (LLM) can perform complex tasks related to digital infiltration and exploitation. The findings highlight a significant leap in the offensive potential of next-generation AI systems.
Advanced offensive capabilities
Researchers focused on the model's ability to assist in the creation and deployment of malicious code. The report indicates that GPT-5.5 can navigate technical hurdles in the cyberattack lifecycle that were previously more difficult for earlier iterations of AI.
While the source material focuses on the parity between OpenAI and Anthropic's products, the institute's findings suggest that the gap in offensive utility between leading frontier models is closing. The institute's evaluation specifically looks at how these models can be leveraged to automate or enhance the scale of cyber threats.
As these models become more proficient at identifying software vulnerabilities, the security industry faces a shift in the landscape of automated exploitation. The institute's research provides a benchmark for the current state of AI-driven cyber threats.
