Spec-Driven vs. Vibe-Coded: John Locke Warns on AI-Created App Security
In a recent Freelock blog post, John Locke critiques the security risks associated with AI-generated applications created through “vibe-coding,” a practice where users prompt large language models to build applications without formal coding.
John references data from SQ Magazine and an analysis by Anthropic to highlight how attackers are using AI to launch cyberattacks, raising concerns about the reliability of code produced by such models. He questions the long-term viability of applications created in this way and contrasts them with more structured approaches like spec-driven development, which incorporate planning, phased implementation, and human oversight.
The post also defends the use of open source platforms such as Drupal and WordPress, noting their established security ecosystems and active maintenance communities. Locke emphasizes that any software exposed to the internet requires ongoing updates and monitoring, and describes Freelock’s use of automated backups, testing, and hotfixes as part of its protection plans for client sites.


