A single cloud compromise could lead to exploitation of AI for disruptive chat services! 🔒

Summary:

Cloud data theft

Organizations that lose access to their cloud data are facing a disturbing trend where cybercriminals are abusing this data to create and resell sexualized AI chat services.
Illegal chatbots

These AI-powered chatbots often use custom jailbreaks to bypass content filters and engage in dark role-playing games, including child sexual exploitation.
Increase in seizures

Researchers at Permiso security noticed a significant increase in attacks on generative AI infrastructure such as AWS bottomespecially as a result of accidentally disclosing credentials online.
Lack of visibility

Many AWS users do not enable logging (which is disabled by default), resulting in a lack of oversight over how attackers use stolen access.
Test case demonstration

To illustrate the problem, Permiso researchers leaked an AWS test key on GitHub with logging enabled. Within minutes, the key was used to offer AI-powered sex chat services, revealing the alarming extent of abuse.
High number of requests

Over two days, the researchers recorded more than 75,000 successful interactions with the AI models, mainly involving sexual content, including references to child sexual abuse.