Lakera Videos - Breaking Point EP3: "Clause AI" - When Legal Docs Become Attack Vectors

[_Val_]

Clause AI is the latest challenge in Lakera’s Breaking Point where we stress-test agentic AI systems that mirror real-world risks.

In this episode, Steve breaks down how every challenge is framed by three pillars:

• Attack Objective – the mission you need to achieve
• Attack Vector – how the attack is delivered
• Attack Scenario – the context that makes the exploit possible

Clause AI highlights what happens when legal research assistants combine multiple tools like retrieval and email, all without the guardrails of least privilege or separation of duties. It’s a timely case study, especially as hype around the Model Context Protocol (MCP) pushes more developers to wire up agentic AI apps at breakneck speed.

🔗 Join the Lakera Momentum community to discuss your own Breaking Point attempts and learn from other players: https://momentum-lakera.slack.com/sig...

[the_rock]

Great video.