Research June 4 6 min read Updating the taxonomy of failure modes in agentic AI systems: What a year of red teaming taught us A surge in real-world attacks against agentic AI systems is reshaping how we think about risk.