// MACHINE LEARNING
AI
Prompt injection, deepfakes, model theft, the EU AI Act, security and policy at the frontier.
- Agentic AI threats: how MCP becomes an attack chain
Agentic AI moves the threat from what a model says to what it does. We map how MCP turns goal hijacking, tool misuse, and privilege abuse into a working attack chain, and the controls that contain it. - MCP security in 2026: the attack surface mappedA technical map of the Model Context Protocol attack surface in 2026: tool poisoning, line jumping, rug pulls, tool-chaining exfiltration, token sprawl, and the RCE flaws that turned MCP servers into entry points.
- Build a secure MCP server in 2026: a hardening guideA practitioner’s hardening guide for MCP servers in 2026: OAuth 2.1 auth, least-privilege tool scopes, sandboxing, egress control, and the tool-definition pinning that blocks poisoning and rug pulls.
- Deepfake vishing 2026: voice-clone fraud explained
Deepfake vishing uses AI-cloned voices to defeat phone-based trust. Here is how voice-clone fraud works in 2026, what it costs, and how to stop it. - Fable 5 and Mythos 5: the US-only gate threat actors will beatWashington cut Fable 5 and Mythos 5 for every foreigner. Meanwhile 34,814 stolen Claude sessions and 74,114 ChatGPT cookies already sit in infostealer logs. A citizenship gate is just KYC, and KYC loses.
- Prompt injection: the 2026 LLM defender’s playbookPrompt injection is now the dominant attack vector against LLM-powered applications — and most teams shipping AI features don’t have a defensive playbook. We map the attack taxonomy, walk through real exploit patterns, and lay out the controls that actually contain the blast radius.
- MCP for WordPress: set up an MCP server in 2026
A step-by-step tutorial for wiring an MCP server into a WordPress site — using the AI Engine MCP adapter — so Claude, Cursor, or any MCP-compatible client can read posts, run admin tasks, and edit content. With auth, scope, and security hardening you actually need. - What is MCP? A 2026 guide to Model Context ProtocolModel Context Protocol (MCP) is the emerging open standard for connecting AI assistants to tools, data, and live systems. This guide explains how MCP servers work, the architecture behind them, and how to build your first one — with security caveats security teams need to know.
- Detecting AI-generated phishing in 2026: a header-forensics, classifier, and DKIM workflowA 2026 workflow for telling AI-generated phishing apart from real correspondence — combining email-header forensics, public LLM-detection classifiers, and DKIM/SPF replay analysis.
- How to host Llama 3 70B locally with Ollama and Open WebUI: a 2026 tutorialA practitioner’s tutorial for running Llama 3 70B locally with Ollama, Open WebUI, and the right hardware. Privacy-sensitive AI work without sending a byte to OpenAI or Anthropic.
- How to red-team your own LLM app: tutorial with Garak, PyRIT, and PromptfooA 2026 tutorial for running structured prompt-injection and jailbreak red-team tests against your own LLM application using NVIDIA Garak, Microsoft PyRIT, and Promptfoo. Open-source, repeatable, CI-friendly.
- What the EU AI Act actually requires from US companies in 2026The EU AI Act’s enforcement window is open in 2026. Here’s what US companies actually need to do, ranked by risk tier and deadline, in plain English.
- How attackers are using AI agents to automate reconnaissance in 2026A practitioner’s look at how threat actors are wiring open-source LLMs and agent frameworks into their reconnaissance pipelines, what that means for defender visibility, and the detection signals that still work.
- Prompt injection attacks: a 2026 field manualA practitioner’s field manual on prompt injection in 2026 — the four attack patterns that work against production LLM apps, the controls that actually mitigate them, and the test cases your red team should be running this quarter.
- How shadow AI is leaking your company’s secrets — and how to find itShadow AI — the AI tools your employees use without IT’s blessing — is the 2026 version of shadow IT, and it’s leaking proprietary code, customer data, and internal strategy at a pace most security teams aren’t measuring.
- The economics of AI agent jailbreaks: who profits when an LLM goes off-railsEvery successful jailbreak prompt has a price. A look at the underground market for AI agent bypasses in 2026 — who builds them, who buys them, and how the profit motive shapes the threat landscape.
- Local AI vs cloud AI: the real security trade-offs in 2026Running models locally feels safer than sending data to OpenAI. Sometimes it is. Sometimes it isn’t. A practitioner’s breakdown of the actual security trade-offs between local and cloud AI deployments.
- AI in the SOC: where it’s actually working in 2026AI in security operations has graduated from vendor demoware to production reality — but only in three specific use cases. Here’s where AI is genuinely changing SOC work in 2026, and where it still doesn’t.
- Learning OpenClaw: Exposing Dangerous DefaultsA practitioner’s account of building a local AI stack with OpenClaw — and discovering that out-of-the-box defaults turn it into a wide-open data exposure surface for prompt injection and remote compromise.
- The EU AI Act: What It Actually RequiresThe EU AI Act, in force since August 2024 and phasing in through 2027, is the first comprehensive AI regulation in any major jurisdiction. Here is what it actually requires, who it applies to, and what organisations should be doing now.
- AI Red Teaming: How to Stress-Test Your AI SystemsRed teaming for traditional software is well-defined. Red teaming for AI systems borrows the term but operates differently. Here is what AI red teaming actually involves, the documented methodologies, and how to structure an effective exercise.
- Open-Source Models vs Closed APIs: A Security ComparisonShould you build on a closed API like GPT-5 or Claude, or run an open-weight model like Llama 4 or Mistral on your own infrastructure? The choice has real security implications that go beyond cost and performance.
- Model Theft and IP: What Happens When Your AI Gets StolenA trained model represents enormous investment in compute, data, and expertise. The threat of model theft — through extraction, distillation, or outright weight exfiltration — is real and increasingly operationalised. Here is the threat landscape and the realistic protections.
- Adversarial Examples: Tricking ML Models with Imperceptible ChangesAdd a small, carefully chosen perturbation to an image and a state-of-the-art classifier sees a school bus instead of a panda. Adversarial examples are the longest-running unresolved problem in machine-learning security and increasingly relevant to deployed systems.
