ChatGPT’s memory feature supercharges prompt injection

Some of the latest, best features of ChatGPT can be twisted to make indirect prompt injection (IPI) attacks more severe than they ever were before.

That’s according to researchers from Radware, who have created a new exploit chain it calls “ZombieAgent,” which demonstrates not merely that ChatGPT is vulnerable to IPI, but that its new connector and memory features can be weaponized to make IPI attacks more persistent and more widespread than previously thought.

“Connectors” are integrations that link ChatGPT with other software platforms (be they mail services, productivity tools, and so on); and it now has long-term memory, allowing information you feed it today to influence its outputs forever after.

Read more at Dark Reading