Researchers warn that a programming style called "vibe coding" is releasing batches of vulnerable code. The teams behind the finding scanned over 43,000 security advisories across the web and found many cases where generative AI tools helped create insecure code. The Vibe Security Radar was built by the Systems Software & Security Lab (SSLab) at Georgia Tech to track these problems.
Graduate research assistant Hanqing Zhao says the radar scans public vulnerability databases, finds the error for each case, and examines code history to see who introduced the bug. If it finds an AI tool's signature, the radar flags the case. So far the tool has confirmed 74 cases, with 14 labelled critical and 25 labelled high. Discovered vulnerabilities include command injection, authentication bypass and server-side request forgery.
Zhao warns that AI models tend to repeat the same mistakes and that developers should review AI output before shipping it to production. SSLab recommends giving AI more detailed prompts and using tools to check generated code for vulnerabilities.
Difficult words
- vulnerability — weakness that allows security problems or attacksvulnerabilities
- advisory — official message about security problemsadvisories
- generative AI — computer systems that create new content
- radar — tool that finds and reports security problemsVibe Security Radar
- flag — mark something to show attention is neededflags
- command injection — attack that runs harmful commands on a system
- authentication bypass — method to access a system without correct login
- scan — look through data to find problems or itemsscanned, scans
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Have you ever used AI tools to write code or text? How do you check the output for mistakes?
- Do you think giving AI more detailed prompts will reduce insecure code? Why or why not?
- What steps could a small development team take to avoid introducing security bugs from generated code?
Related articles
People learn to use robotic leg prostheses but misjudge their gait
A four-day study found that people who practised with a robotic lower‑limb prosthesis improved their walking but misjudged their own movement. Researchers say better visual feedback could help users calibrate their body image and gait.
Australian creators launch 'Stop AI Theft' campaign
Artists, journalists and Aboriginal cultural workers in Australia have launched the "Stop AI Theft" campaign. They say generative AI uses their work without permission, harms jobs and income, and they demand payment, rights and transparency.
New training method helps models do long multiplication
Researchers studied why modern language models fail at long multiplication and compared standard fine-tuning with an Implicit Chain of Thought (ICoT) method. ICoT models learned to store intermediate results and reached perfect accuracy.
Instagram bot campaigns target activists and media in Western Balkans
In November 2025 coordinated bot operations hit activist and media Instagram accounts in the Western Balkans. Reports on November 16 and 23, 2025 describe fake followers, mass reports, bot comments and mass liking that reduced visibility.