AI models can’t fully understand security – and they never will

Despite the hype around AI-assisted coding, research shows LLMs only choose secure code 55% of the time, proving there are fundamental limitations to their use.