Designing AI agents to resist prompt injection

a16z leads a $20M seed into Runta at a $100M+ valuation for its agent execution layer — sandboxes, spending caps and audit trails that treat agents like children who need supervision. Martin Casado called it 'core systems software rebuilt for agents'; the round came after prompt-injection containment failures made agent governance urgent.

How OpenAI uses chain-of-thought monitoring to study misalignment in internal coding agents—analyzing real-world deployments to detect risks and strengthen AI safety safeguards.