Study Finds Severe Security Flaws in OpenAI, Anthropic AI Coding Tools

• Study finds AI coding tools from OpenAI and Anthropic produce vulnerable websites easily hacked
• Only 10.5% of AI-generated code deemed secure in strong security setups
• Anthropic blocks competitors from accessing Claude models while expanding into healthcare

The AI coding revolution promised to democratize web development, but apparently nobody told the hackers they weren't invited to the party. Tenzai's study reveals that popular AI tools are churning out websites with more holes than a golf course, complete with vulnerabilities that would make a 1990s PHP developer blush. When your AI assistant can build an e-commerce site that practically hands over credit card data to anyone who asks nicely, we might need to reconsider our definition of "artificial intelligence."

The root problem lies in AI training data that includes decades of questionable coding practices, creating a digital inheritance nobody wants. These models excel at speed and functionality but treat security like an optional garnish rather than the main course. As businesses rush to adopt AI for rapid prototyping, they're essentially playing Russian roulette with their users' data, except the chamber is mostly loaded and the gun is pointed at everyone's bank account.

Meanwhile, Anthropic is playing corporate chess by blocking competitors while expanding into healthcare, where security flaws could literally be life-threatening rather than just wallet-threatening. The irony is delicious: as AI companies fight over market share, they're simultaneously creating the very problems that could undermine their entire industry. Perhaps the real artificial intelligence was the security vulnerabilities we made along the way.