Open-Source AI Models Vulnerable to Criminal Misuse, Warn Researchers

Researchers from SentinelOne and Censys found that open-source AI models, like those based on Meta's Llama, can be easily misused by hackers.

These models, which are often run on personal computers, can be used for harmful activities like creating spam, phishing content, or spreading disinformation.

The researchers studied these models for 293 days and found that many of them have removed safety features, making them more vulnerable.

They also found that a significant number of these models are being used in China and the US.

Experts say that while the creators of these models are not responsible for every misuse, they should still take steps to prevent foreseeable harm.

Companies like Meta and Microsoft have tools and guidelines to help developers use these models responsibly.