Affiliate links fund independent research. All rankings and technical analysis remain strictly independent.

AI Threats, Privacy & Analysis

Long-form briefings on AI-powered threats, privacy defence, and defensive guidance, with citations and practical context for technical teams.

Security Engineering

The Zero-BS Guide to Preventing Prompt Injection Attacks

A tactical guide to mitigating prompt injection attacks in production. Moving beyond fragile regex filters to semantic validation and dual-LLM architectural defenses.

6 minJosh Cabana
AI Threats

Claude Extension Flaw Enabled Zero-Click XSS Prompt Injection via Any Website

A critical vulnerability in Anthropic's Claude Chrome Extension exposes users to zero-click XSS prompt injection attacks, allowing malicious actors to execute commands without user interaction. This article examines the technical details, risks, and defensive strategies for organizations and individuals.

5 minJosh Cabana
AI Threats

Security Researchers Sound the Alarm on Vulnerabilities in AI-Generated Code

A surge in vulnerabilities linked to AI-generated code has prompted security researchers to call for urgent improvements in code review and development practices. This article explores the risks, trends, and defensive strategies for organizations leveraging AI in software development.

5 minJosh Cabana
AI Threats

AI Flaws in Amazon Bedrock, LangSmith, and SGLang Enable Data Exfiltration and RCE

Three independent research teams disclosed critical vulnerabilities across Amazon Bedrock AgentCore, LangSmith, and SGLang in March 2026 — collectively enabling DNS-based data exfiltration, account takeover, and unauthenticated remote code execution against AI infrastructure. One of these flaws remains unpatched, and CERT/CC has issued a public advisory.

7 minJosh Cabana
AI Threats

CursorJack: How MCP Deeplinks Turn Cursor IDE Into a Code Execution Vector

Proofpoint Threat Research has demonstrated that a single crafted deeplink can weaponise Cursor IDE's MCP installation flow to execute arbitrary commands under a developer's full privileges — exposing the structural security gap at the heart of the Model Context Protocol ecosystem.

6 minJosh Cabana
AI Threats

LLM Guardrails Are Failing: What the 2025–2026 Research Actually Shows

Palo Alto Networks' Unit 42, Oxford researchers, and a Nature Communications study converge on the same finding: the safety layers enterprises rely on to govern generative AI can be bypassed at scale, by automated tools, in under 60 seconds. The attack success rates are not marginal — they are, in several documented cases, approaching 100%.

8 minJosh Cabana
AI Threats

Hive0163 Deploys AI-Assisted Slopoly Malware for Persistent Access in Ransomware Attacks

IBM X-Force discovered Slopoly, an LLM-generated PowerShell backdoor deployed by Hive0163 that maintained persistent access for 7+ days during a live ransomware engagement. The code self-describes as a "Polymorphic C2 Persistence Client" — but isn't actually polymorphic. That gap between what AI claims to build and what it actually builds is the story.

7 minJosh Cabana
AI Threats

OpenClaw AI Agent Flaws: CVEs, Prompt Injection, and a Government Warning

China's CNCERT issued a formal warning on March 14 about OpenClaw's inherently weak default configurations. With 97+ CVEs tracked, 135,000+ exposed instances, and zero-click data exfiltration demonstrated in the wild, the open-source AI agent has become a case study in what happens when autonomous systems ship without a security baseline.

8 minJosh Cabana
Privacy

Australia's Privacy Act Reforms 2026: What You Need to Know

Australia's Privacy Act has undergone its most significant overhaul since 1988 — with new penalties reaching $50 million, mandatory disclosure of AI decision-making systems, and a new statutory tort for serious privacy invasions. Here's what changed and what it means for your organisation.

7 minJosh Cabana
AI Threats

Agentic AI Security Risks: What Every Developer Must Know

Autonomous AI agents are now embedded in enterprise workflows with privileged access to databases, APIs, and critical systems — but the security infrastructure governing them hasn't kept pace. Here's what developers and security teams need to understand right now.

7 minJosh Cabana
AI Threats

AI Model Prompt Injection Attacks Explained

Prompt injection is the #1 vulnerability in deployed AI systems — and unlike traditional software flaws, it cannot be patched. Here's how direct and indirect attacks work, real-world exploits from 2024–2026, and the defence strategies that actually reduce risk.

7 minJosh Cabana