Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Here's the uncomfortable part first: in August 2026, researchers found AI agents connected to Hugging Face running loose inside enterprise networks. No owner, no audit trail, nobody who could tell you they existed until something broke. If that sentence made your stomach drop a little, good, because it should. It's the same blind spot most security teams are sitting on right now. They just haven't had their version of the incident yet.

The volume of published incident research involving agentic AI is increasing, and the analysis that follows each report tends to concentrate on the same attribute: speed. The recent investigation from Unit 42, the threat intelligence and incident response group at Palo Alto Networks, is a representative case.

In July 2026, an OpenAI model escaped its evaluation sandbox and broke into Hugging Face's production infrastructure. It is the first documented end-to-end intrusion carried out by an autonomous AI agent. The most repeated takeaway, "the AI went rogue," is also the least useful one. The real lessons are about containment engineering, about who is allowed to use powerful models, and about why the coming wave of regulation could easily leave defenders weaker than attackers.

Editor's note: This article was originally published by Tim Erlin on LinkedIn. It has been republished here with the author's permission. It seems like every security vendor now sells "AI security." The WAF companies, the API gateway companies, the cloud platforms, the proxy startups: all of them have an AI story, and most of them have attached one of three labels to it. AI gateway. AI firewall. AI control platform. The terms often get used as if they're interchangeable, but they are not.

Every week, someone in your organization stands up an AI service. Maybe they told security about it, but probably not. By the time it shows up in your inventory, it has been running for weeks, processing data, calling external APIs, and doing things nobody formally reviewed.

Editor's note: This article was originally published by Craig Riddell on LinkedIn. It has been republished here with the author's permission. Boards are giving AI security more airtime than ever. What they're not giving is the right framing. A year or two ago, AI was mostly a question of experimentation risk. Today, it's tied directly to revenue, customer experience, operational efficiency, and competitive advantage. The urgency is real, and it's translating into aggressive deployment timelines.

The Model Context Protocol (MCP) is a de facto standard for providing structured access to privileged systems for AI agents and external integrations. It acts as a USB-C port for AI, enabling faster innovation by allowing organizations to expose tools, resources, and workflows without the time-consuming work of building APIs. Adoption has surged in recent months, and categories like payments, project management, and developer platforms are already beginning to reap the benefits.

Most organizations treating AI security as a model problem are defending the wrong layer. Security teams filter prompts, patch jailbreaks, and tune model behavior, which is all necessary work, while the actual attack surface sits largely unexamined underneath. That surface is the API layer: the endpoints AI systems use to retrieve data, call tools, and take action on behalf of users. This isn't a theoretical gap.

Every secure API draws a line between code and data. HTTP separates headers from bodies. SQL has prepared statements. Even email distinguishes the envelope from the message. The Model Context Protocol (MCP), the fast-growing standard for connecting AI agents to external services, inherits that gap from the models it sits on top of.

Your legal team just handed you a 400-page document and said "figure out compliance." The EU AI Act is live, your organization falls under its scope, which is broader than many expect. Even non‑EU companies must comply if their AI systems are used, deployed, or produce effects within the European Union. In practice, that means that global organizations building or integrating AI models cannot treat the Act as a regional regulation.