ThreatsDay Bulletin Highlights Diverse Cyber Threats: AI Hijacking, Data Breaches, and Vishing Operations

The cybersecurity landscape continues to evolve rapidly, with the latest ThreatsDay Bulletin showcasing a diverse array of threats ranging from advanced AI agent hijacking to significant data breaches and large-scale vishing operations. This week's updates underscore the persistent need for vigilance across various sectors, highlighting new attack vectors and the exploitation of existing vulnerabilities.

A novel attack dubbed 'GhostJacking' expands upon the 'Agentjacking' concept, specifically targeting AI agents. This method tricks AI agents into executing arbitrary code on developer machines by leveraging their trusted access. The attack can be initiated through seemingly innocuous inputs like poisoned logs or alerts, allowing attackers to escalate privileges, pivot to enterprise cloud infrastructure, exfiltrate data, and establish persistence. Researchers emphasize that traditional defenses are often ineffective as the AI agent's actions remain within its legitimate permissions, highlighting the critical need for robust guardrails around AI agents.

In the realm of data breaches, ShipMonk, a shipping provider for Trezor, has experienced a significant incident exposing sensitive customer order data. This breach potentially impacts customers in the U.S., U.K., Sweden, Colombia, Brazil, Italy, and Portugal who received orders between May 10 and August 8, 2026. The exposed information includes full names, shipping addresses, phone numbers, and email addresses, posing a risk of further phishing or identity theft.

Cursor, a coding agent provider, has addressed a critical vulnerability in its command-line interface (CLI) that allowed cloned repositories to execute arbitrary commands on a developer's machine without explicit consent, even when sandbox protections were enabled. The flaw, discovered by Manifold Security, resided in the .cursor/worktrees.json file and could lead to unauthorized access to sensitive system resources, including SSH keys and cloud credentials. A patch was released shortly after responsible disclosure.

Okta has shed light on 'Work Panel,' a sophisticated operator console used by threat actors to orchestrate large-scale vishing campaigns targeting identity providers. This platform streamlines the process of registering phishing domains, cloning brand assets, and launching new phishing sites within minutes. The multi-tenant nature of Work Panel indicates a well-established cybercrime ecosystem supporting these voice phishing operations, with UNC6671 (also known as Cordial Spider) identified as a user of the tool.

Additionally, the bulletin details an ongoing campaign named 'City-Forum' that targets unauthenticated guest user access in Salesforce Experience Cloud sites and ServiceNow Service Portals. Threat actors are exploiting lesser-known techniques, including the UI-API for Salesforce and a native ServiceNow search endpoint, to exfiltrate data. The targets include major industries like telecommunications, banking, and enterprise software vendors, with one victim recording over 560,000 suspicious events from the attacker's IP address.

Meta has also introduced an optional 'Scam Alert' feature for WhatsApp, utilizing an on-device machine learning model to identify and warn users about potential scam messages without sending message content off the device. This move aims to enhance user safety by proactively detecting malicious communications directly on the user's phone.

These diverse threats, from AI agent manipulation to traditional data breaches and large-scale phishing operations, collectively paint a picture of a dynamic and challenging threat landscape. Security teams must remain adaptable, focusing on securing AI integrations, protecting customer data, and implementing robust defenses against evolving social engineering tactics.