Build your own vulnerability harness

Anthropic's security research demonstrates that effective vulnerability detection requires a model-agnostic "harness" architecture that treats AI models as interchangeable components rather than relying on a single frontier model. The system uses multiple models in sequence for cross-validation, traces vulnerabilities across interconnected repositories, and maintains persistent state across runs—capabilities that isolated AI agents cannot provide. This approach protects against market shifts where leading models become unavailable or obsolete while ensuring comprehensive security coverage through diverse analytical perspectives.