A new Wormable vulnerability discovered in GPT-3

Security researchers have identified a new concerning vulnerability in GPT-3 models, known as WormGPT. This vulnerability allows an attacker to execute a wormable attack, meaning that malicious code can automatically spread to other vulnerable systems without the need for human interaction.

The WormGPT vulnerability exploits the ability of GPT-3 models to generate code in programming languages. Researchers have found that by feeding GPT-3 with example code, the model can continuously generate similar code, allowing an attacker to create a computer worm that spreads automatically.

This type of wormable attack is particularly dangerous as it can quickly spread to other network-connected systems, causing widespread damage. WormGPT can be used to carry out denial-of-service attacks, steal confidential information, and even take control of compromised systems.

The researchers have reported their findings to OpenAI, and the company has acknowledged the severity of the vulnerability and is already working on a fix.

Source: gbhackers