As frontier models slap on more classifiers to prevent cyber abuse, attackers and defenders are forced to switch to open weight models…

AI, Social Engineering, Prompt Injection, LLM Data Leakage

The Chinese threat group Jewelbug has repurposed a single piece of infrastructure to serve two distinct objectives simultaneously: surveilling government organizations and profiting from cryptocur…

At COMPUTEX 2026, Qualcomm officially unveiled its new Snapdragon C computing platform, purpose-built for entry-level Windows laptops. The platform targets students, families, small businesses, an…

Google recently announced a significant policy adjustment regarding artificial intelligence. In the future, users generating images, videos, or music with Google’s AI models can manually rem…

Researchers at watchTowr published a working exploit for a Citrix NetScaler RCE flaw this week. The bug, CVE-2026-8452, lets an unauthenticated attacker run code as root on the appliance. watchTow…

Microsoft is currently developing a customizable right-click context menu for Windows 11. Screenshots of the feature surfaced online, revealing that the new menu applies to both the desktop and Fi…

To comply with the transparency regulations stipulated by the European Union Artificial Intelligence Act (EU AI Act), Anthropic recently disclosed the underlying mechanics of its text watermarking…

As Google officially began hosting third-party app marketplace applications, including Aptoide, on the US Google Play Store this week, the years-long battle for an open ecosystem appears to have r…

Welcome to the latest CVE WATCHTOWER briefing. Our team compiled this weekly threat intelligence report to highlight active cyber risks. Between August 10 and August 16, 2026, security researchers…

From manual QA tester to writing my own recon and security tooling — a practical breakdown of what I built, why, and what surprised me.

30-Second Version: The most-upvoted account takeover in HackerOne’s entire disclosed-reports dataset contains no exploit, no injection, no…

In July, during a cybersecurity test, one of OpenAI’s autonomous AI agents stopped treating the test as a boundary.

Summary:

One field in a JWT header. That’s all it took to slip past 8x8’s API.

During a security assessment on a popular e-commerce platform, I stumbled upon a classic yet frequently overlooked vulnerability, an…

Role-based access control often begins as a small feature: administrators can manage users, operators can process transactions, and…

It is commonly believed that “hacking” is a complex technical attack in which hackers choose passwords and penetrate systems.

Cyber threats are becoming more advanced, and organizations of every size are exposed to risks through websites, applications, networks…

Modern web applications increasingly depend on APIs to connect frontends, mobile apps, microservices, third-party platforms, and internal…

Why AI applications need a different approach to application security.