Blog
- 40 months of prompt injection.
A timeline of prompt injection research, attacks, and mitigations from the first demonstrations in late 2022 through the beginning of 2026.
- There Is No Firewall for English.
Agentic tooling introduces a new class of security risk: documents with no malware that can trigger malicious behavior when processed by an AI assistant.
- The Webpage Has Instructions. The Agent Has Your Credentials.
Prompt injection is the most critical agent security threat. How attackers hijack agents via webpages, MCP metadata, and tool outputs, and how to defend.
- The Wiring Is More Dangerous Than the Weights.
As agents grow autonomous, the real risk shifts from model weights to the interfaces, protocols, and permissions connecting them to the world.
- What Is OpenGuard?
OpenGuard is a local proxy that sits between your coding agent and its LLM provider, letting you inspect and control every request and response.