Another Day, Another AI Prompt Injection - Threat Wire

Quick Overview

The video discusses a new "Prompt Injection" vulnerability affecting Google's Gemini AI, allowing attackers to hijack conversations or extract sensitive information by embedding malicious commands into calendar invites or other data. It also touches upon the ineffectiveness of current phishing training methods and a Salesforce data breach.

Key Points: A "Prompt Injection" vulnerability has been found in Google's Gemini AI, allowing attackers to manipulate its behavior via crafted calendar invites. Research presented at Black Hat USA 2025 suggests traditional anti-phishing training and embedded phishing simulations have limited effectiveness. A significant Salesforce data breach, linked to the "ShinyHunters" group, has impacted major companies like Qantas, LVMH, and Adidas. The Salesforce breach involved voice phishing and social engineering to steal data from CRM instances. Google itself was affected by the Salesforce breach, leading to updates in its file access protocols to prevent future attacks. Microsoft is updating its M365 security settings to disable certain file access protocols by default, starting July 2025. Hack Five has released a new product, the "pineapple bee-per," designed for various functionalities.

Context: This video from "Threat Wire" covers recent cybersecurity threats and vulnerabilities. It begins by discussing a prompt injection flaw in Google's Gemini AI, then shifts to the questionable effectiveness of current employee phishing training, and finally addresses a major data breach at Salesforce attributed to the "ShinyHunters" group. The content highlights the evolving landscape of cyber threats and the challenges organizations face in protecting sensitive data.

Detailed Analysis

The "Threat Wire" video details a newly discovered "Prompt Injection" vulnerability that targets Google's Gemini AI. This exploit allows attackers to manipulate Gemini's behavior, potentially hijacking conversations, extracting sensitive information, or controlling other AI agents. The vulnerability is demonstrated by crafting malicious calendar invites that, when processed by Gemini, can execute harmful commands. The video also briefly mentions research from Black Hat USA 2025 that found traditional anti-phishing training and embedded phishing simulations to be largely ineffective in preventing users from falling for phishing attempts. Furthermore, it highlights a significant Salesforce data breach attributed to the "ShinyHunters" group, which used voice phishing and social engineering tactics to steal data from Salesforce CRM instances, impacting major companies like Qantas, LVMH, and Adidas. The segment concludes with a mention of Hack Five releasing a pineapple bee-per, a device that can be used for various tasks, including potentially malicious ones.