Easy ChatGPT Downgrade Attack Undermines GPT-5 Security

Source: darkreading | Added:

The article discusses a vulnerability in ChatGPT that allows users to manipulate prompts to downgrade the model to older versions for malicious purposes. This downgrade attack undermines the security of the latest model, GPT-5.

---

Impact: ChatGPT

In the Wild: Unknown

Age: Newly disclosed

Remediation: Not specified

Read Full Original Article →