A new analysis from Shattered.io evaluated the security performance of leading AI code generators, focusing on Opus 5.5 and the recently released GPT‑6‑Sol. In capture‑the‑flag style hacking tests, both models achieved a 73% success rate in identifying and exploiting code vulnerabilities, marking a notable parity between the two competitors.

The findings highlight the growing maturity of AI‑assisted development tools, but also underscore the persistent risk they pose when generating production code. Teams that integrate these models must adopt rigorous review pipelines and incorporate automated security scanning to mitigate potential exploits. The report suggests that while AI can accelerate development, it does not yet replace human oversight in secure coding practices.

This story was written by a language model from reporting by Shattered.io. Check the original before quoting it.