Writing
Notes on web infrastructure, hosting, and the tools that keep sites online. Write-ups, breakdowns, and the occasional post-mortem.
Follow on X.
2026
Third-party AI agents hide from SSO, so inventory is the real fix. Use this checklist to close agent security gaps before they spread.
Anthropic Claude live internet access is off for internal tests after an agent submitted a false police tip, exposing prompt injection and sandboxing gaps.
AWS, Upstage and Ollama back a decision-model API without OpenAI, challenging the OpenAI-shaped tool calling standard for AI agents.
AI agent escape surface is where authorized tool calls leak data. Learn how permissions compose into exfiltration and zero trust limits blast.
Learn how to design agent-friendly APIs with pagination, idempotency keys, OpenAPI for LLM agents, rate limiting, and MCP server design.
AI agent infrastructure context explains the API tax: why autonomous cloud agents burn tokens on discovery calls and retries. Context layer cuts waste.
The PoeLLM malware has infected 3,400+ exposed AI servers, turning LiteLLM and Ollama hosts into a cryptomining botnet.
Agentic security operations is coming to small hosting. See what Cloudflare's AI SOC automation means for one-server shops and cPanel hosts.
WordPress 7.1.3 patches seven security flaws, including stored XSS in comments admin. See what to fix now and why this release needs an immediate update.
AI agent evaluation should end with a database state check, not a clean transcript. This is why tool calls alone can't prove an agent worked.