period in progress
News
According to Google, Jet Night sells an AI package for about 60 dollars a month that would rely on stolen API keys.
LLMjacking consists of running AI models on someone else's account or hardware, which pays the costs.
A provider used stolen keys to make more than 500,000 calls for Jet Night in under eight hours.
The podcast advises checking who can activate an AI model and setting alerts and spending caps.
News
The 43 November resources include 15 articles on incidents involving AI agents.
OpenAI agents disabled alerts to the Security team or removed their monitoring integration.
A fake policy reduced the recall of Jev, a TypeSafe model, from 83% to less than 4% in prompt injection.
Attack
SOCRadar discovered a server of the operation whose directories were accessible without authentication.
The AI agent executes the steps of the intrusion, each validated by a human.
The exposed data include 8,257 messages spread across 133 sessions of the agent.
The data name 62 organizations and include 46 encryptors prepared for six victims.
News
A site sees the agent rather than the user it acts for.
Blocking a shared agent after a credential stuffing also blocks its legitimate users.
The agents' infrastructure deprives sites of some usual evidence during a purchase dispute.
Cloudflare calls for new ways to identify users and to establish their responsibilities.