Download counts aren't trust: a lesson for AI coding agents
A malicious copy of a popular maths library reached six million downloads in two days. Why agents that pick packages are exposed, and what limits the damage.
3 min read
Threats, controls and secure design for AI tools and the systems they touch.
A malicious copy of a popular maths library reached six million downloads in two days. Why agents that pick packages are exposed, and what limits the damage.
3 min read
NIST summarised industry responses on AI agent security and is building control overlays for agents. The themes, the status, and steps to take now.
3 min read
The seven checks a connection passes before it reaches a SpanDock server, which ones stop a stranger or another customer, and how to try it in the lab.
4 min read
The AI Omnibus moved the EU AI Act's high-risk deadlines and simplified AI literacy. The new dates, what enforcement looks like, and what to do now.
3 min read
Claude Code and Codex both ship operating-system sandboxes for agent commands. What they restrict, what runs outside them, and settings worth enforcing.
4 min read
Advisories from Splunk, AWS, Google and the Rust MCP SDK share one pattern: a tool builds a request or command from input it shouldn't trust.
Add two or three SpanDock guardrails for secrets, personal data and your own patterns, enforce them on every computer, and check that they work.
Flowise, Langflow and PraisonAI fixed flaws where prompt injection led to code execution. The shared mistake, and what to check in your own agents.
4 min read
Give every computer's AI tools one governed MCP server: add two or three connectors read-only, set permissions and approvals, and check it works.
5 min read
MCP 2026-07-28 drops sessions, adds routing headers, hardens authorisation and deprecates features. What platform and security teams need to plan for.
5 min read