ai freeup

A tool description card with hidden instructions concealed behind its visible summary, illustrating how an MCP tool poisoning attack reaches an AI agent

How an MCP Tool Poisoning Attack Works

August 14, 2026

An MCP tool poisoning attack hides instructions in the tool metadata your agent reads and you never see. How a tool you already approved can change underneath you, and what to check first.

Vetting AI vendors: a vendor icon whose reach extends unseen into unrelated third-party systems

What the OpenAI-Hugging Face Sandbox Escape Means for Vetting AI Vendors

August 12, 2026

OpenAI's own AI agent escaped its sandbox, breached Hugging Face, and reached four accounts on unrelated services neither firm agreed to expose. What vetting AI vendors should ask before the next one.

A narrow, scoped set of keys handed to an AI agent instead of a full keyring, illustrating how to limit AI agent permissions

How to Limit AI Agent Permissions (Before It Needs You To)

August 10, 2026

How to limit AI agent permissions in practice: scope access at the source, add a checkpoint before actions run, isolate execution, and keep a record. Includes where to start today.

An AI agent taking real actions like sending email and editing files on its own, illustrating AI agent security risks

AI Agent Security Risks: When AI Acts on Its Own

August 9, 2026

AI agent security risks explained in plain English: once AI can act on its own, sending email or changing data, a small mistake becomes a real one. Here's what can go wrong and why it matters.

A developer machine quietly running several unapproved shadow MCP servers that no inventory or security tool can see

Shadow MCP Servers: The AI Tools Nobody Vetted

August 4, 2026

Shadow MCP servers get connected in seconds, by developers, through config files nobody reviews. Here is why your usual discovery tools miss them and how to find what is already running.

An AI assistant connected to several MCP servers that reach into files, databases and email, illustrating MCP server security risks

What Is MCP? MCP Server Security Risks Explained

August 3, 2026

MCP server security risks come from one design choice: the server tells the AI what it can do, and the AI believes it. Here is what MCP actually is and where that trust quietly breaks down.

A terminal window where an AI coding agent runs shell commands, illustrating the risk of securing AI coding agents

Securing AI Coding Agents: Shell and Network Access

August 2, 2026

Securing AI coding agents is a different problem from securing a chatbot: they run shell commands and reach the network. Here's what goes wrong in practice and the guardrails that actually contain it.

Direct vs indirect prompt injection: a user typing a malicious prompt versus a hidden instruction buried in a document

Direct vs Indirect Prompt Injection: Simple Examples

July 29, 2026

Direct vs indirect prompt injection, explained with simple examples: one is typed straight into the chat, the other hides in web pages, resumes, and PDFs your AI reads. Here's how to tell them apart.