← back
Spec-Driven Testing for Agents With A Brain the Size of A Planet — Steven Willmott, SafeIntelligence
Takeaway
Specify rules, roles, domain meaning, and robustness requirements explicitly before generating tests for an agent.
Summary
- A more capable model is not automatically safer: it may understand sophisticated malicious instructions and have a broader surface of possible actions.
- Spec-driven validation defines the agent's task and role independently of its implementation, going beyond a dataset of good input-output examples.
- Specifications should capture business rules, ontologies, domain terminology, permissions, and expected behavior under different user roles.
- Robustness testing checks stability under typos, rephrasing, and other perturbations, while tool authority and task scope determine potential harm.
spec-driven-testingagent-validationrobustness
Original description
Wrapping a malicious instruction in a poem is an effective jailbreak against large models and not against small ones. Small models don't understand the poem. Large models do and execute the instruction. Steven Willmott from Safe Intelligence argues this is one reason bigger is not straightforwardly safer: a larger model with broader capabilities has more attack surface and more infrastructure access to abuse. His frame is spec driven validation. An agent spec is not just a test dataset. It needs explicit rules (never offer more than 10% discount), domain ontologies (an airline agent only needs to know about destinations that airline actually flies to), rights and roles, and robustness requirements such as how many typos or rephrasings before it fails. Write these independently of the implementation so they survive a model swap and can drive both security testing and iterative improvement. Speaker info: / stevenwillmott https://x.com/njyx