Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

From inchoate brainstorming sessions in the halls of Dartmouth College to a panoply of funding springs and winters, AI has made its way into the tech stack of not just almost every enterprise but also every household. This, though music to the ears of an AI researcher, rewards a security professional with sweat beads. Even a single AI/ML/LLM or an MCP feature in your product evolves your attack surface, necessitating scouting for the right AI/ML/LLM/MCP penetration testing companies.

A Russian-speaking threat actor known as “Trim” has spent the better part of 2026 systematically dismantling the guardrails on publicly available frontier AI models and rebuilding them as offensive tools. What started in March as a knowledge-sharing post on a Russian cybercrime forum detailing how to break Claude Opus into writing malware, had evolved by June into a fully productized, commercially marketed AI-powered penetration testing platform.

Claude is now a part of many employees’ everyday work. But even as a sanctioned application, it creates a visibility problem for security teams. Sensitive data can move into prompts, files, projects, and conversations, and teams need a practical way to see what happened and whether it violates policy.

Shadow AI is the fastest-growing unmanaged risk surface in most organizations. Employees are adopting AI tools through browser extensions, free-tier SaaS accounts, personal logins, and embedded platform features without involving IT, security, or procurement. The result is an expanding footprint of AI systems that process corporate data, generate business outputs, and create compliance exposure while remaining invisible to the governance program responsible for managing those risks. ‍

Cyber risk benchmarking is the practice of measuring an organization's security posture, quantified exposure, and operational metrics against comparable companies in the same sector and size band. Done well, it answers three questions a board expects the CISO to answer. ‍ ‍ Done poorly, it produces vanity metrics that look impressive in a slide deck and mean nothing when the auditors, regulators, or insurance carriers start asking questions. ‍

Learn how to measure time to revoke for exposed credentials using validation and invalidation timestamps, remediation SLAs, and CISO reporting metrics.

The first time an MCP (Model Context Protocol) server felt real to me, it wasn't because of a clean demo. It was because of the noise. TL;DR: The harness matters more than the protocol, and the evidence matters more than both. MCP earns its keep when it shortens the path from a good security question to trustworthy evidence, and almost everything interesting about making that work happens in the harness wrapped around the model. In this series, I will cover how to build an MCP for an AI SOC.

In a previous post we looked at the skills CLI by vercel-labs/skills and ways for a malicious skill to overwrite an existing trusted skill by using homoglyph names or abusing weird CLI behaviors. This post will delve into an OSC-8 escape injection we found in the skills add command, which lets a skill author write arbitrary text to the console and clickable links as if it's part of the CLI's output.

Attackers do not need to defeat every control. They only need to find one account, system, or recovery process that falls outside the policy.