CyberSecEval produces security findings that need to be traceable and tamper-evident — especially when results feed into compliance decisions or when multiple teams need to verify the same assessment.
Signet could add a cryptographic layer to PurpleLlama's output:
- Sign each evaluation result — Ed25519 signature proves who ran the assessment, when, and that the output hasn't been modified
- Hash-chain the assessment log — sequential evaluation steps are linked cryptographically, so reordering or deleting results is detectable
- Offline verification — anyone can verify the assessment's integrity without contacting a server
This matters when CyberSecEval results are used as evidence in security audits or when assessment pipelines run across teams that don't share infrastructure.
Signet is available as pip install signet-auth (Rust core via PyO3). A minimal integration would be a wrapper that signs each benchmark result and appends to an audit log:
from signet_auth import SigningAgent
agent = SigningAgent.create("cyberseceval-runner")
receipt = agent.sign("cyberseceval.run", params={"benchmark": "instruct", "model": "llama-3"})Apache-2.0 / MIT dual licensed. No SaaS dependency.
CyberSecEval produces security findings that need to be traceable and tamper-evident — especially when results feed into compliance decisions or when multiple teams need to verify the same assessment.
Signet could add a cryptographic layer to PurpleLlama's output:
This matters when CyberSecEval results are used as evidence in security audits or when assessment pipelines run across teams that don't share infrastructure.
Signet is available as
pip install signet-auth(Rust core via PyO3). A minimal integration would be a wrapper that signs each benchmark result and appends to an audit log:Apache-2.0 / MIT dual licensed. No SaaS dependency.