The Oort Blog.

Thoughts, research, and updates on the future of autonomous AI infrastructure.

Your AI Agent Can Browse the Web. So Can Attackers.
·8 min read · Updated Mar 29, 2026

Your AI Agent Can Browse the Web. So Can Attackers.

Browser-capable AI agents face prompt injection from web content, credential theft, and sandbox escapes. Here's the attack surface most teams aren't ready for.

Oort Labs Team arrow_forward
Introducing Silo: A Secure Runtime for AI Agents in Production
·4 min read · Updated Feb 27, 2026

Introducing Silo: A Secure Runtime for AI Agents in Production

AI agents are moving to production, but they need more than prompt guardrails. Discover Silo, a secure execution runtime for autonomous AI workloads.

Oort Labs Team arrow_forward
Exposed AI servers: how LLMjacking happens (and how to stop it)
·12 min read · Updated Feb 20, 2026

Exposed AI servers: how LLMjacking happens (and how to stop it)

Running Ollama, vLLM, or an MCP server? Here’s how self-hosted LLMs end up exposed, what attackers do, and the practical hardening steps that reduce risk.

Oort Labs Team arrow_forward
Prompt injection is the new SQL injection (and why prompt-only defenses keep failing)
·8 min read · Updated Feb 13, 2026

Prompt injection is the new SQL injection (and why prompt-only defenses keep failing)

Prompt injection is the new SQL injection for LLM apps. Learn how attacks work, why common mitigations fail, and what actually reduces blast radius.

Oort Labs Team arrow_forward
AI agents in 2025: what actually happened vs the hype
·12 min read · Updated Feb 6, 2026

AI agents in 2025: what actually happened vs the hype

2025 was supposed to be the year AI agents went mainstream. The investment was real, the ambition was real - but so was an 80% enterprise failure rate. Here's what the data actually shows.

Oort Labs Team arrow_forward