DUBLIN, Calif. — April 8, 2026 — Safeguard today introduced a three-model artificial intelligence lineup engineered specifically for software supply chain security: Griffin for deep reasoning, Eagle for ranking and triage, and Lion for on-device inline inference. Each model in the lineup is trained on a curated cybersecurity-only corpus, with no general-purpose web crawl in the pretraining data.
The corpus thesis underlying the lineup is direct: a model that has only ever read advisories, exploit write-ups, signed SBOMs, VEX statements, taint analyses, compliance frameworks, and code patches will produce more faithful security reasoning than a general-purpose model fine-tuned after the fact. Safeguard built the data pipeline, the training stack, and the evaluation harness around that premise.
The Griffin reasoning family ships in five sizes — Griffin Zero (671B-MoE), Griffin L, Griffin M, Griffin S, and Griffin Lite — allowing customers to match model capability to deployment envelope. Griffin Zero is intended for sovereign and air-gapped regulator workloads; Griffin Lite is small enough to run alongside a CI runner. All five variants share the same instruction format, evaluation suite, and policy interface, so customers can promote workloads up the family without re-training prompts.
Eagle is a dedicated ranking head used to order findings by exploitability, blast radius, and reachability. Rather than returning a flat list of CVEs, Eagle produces a ranked queue with cited evidence — the call path, the affected function, the reachable sink, and the upstream advisory. Eagle is invoked by the platform's triage layer and is also exposed as a standalone scoring API for customers integrating with existing ticketing systems.
Lion is the on-device inline model. It runs inside the developer's editor or inside an MCP-governed tool boundary and is responsible for sensitive-data egress guardrails, inline secret detection, and policy-bounded autocompletion. Lion is small enough to run on a laptop and is the only member of the lineup that operates outside the central reasoning pipeline.
"We did not want to ship a security product wrapped around a general model," said Hritik Kumar Sharma, Founder and CEO of Safeguard. "If the corpus is mostly Reddit and Wikipedia, the reasoning will reflect Reddit and Wikipedia. Griffin, Eagle, and Lion are what you get when the corpus is, instead, every advisory, every signed SBOM, and every well-formed taint analysis we could ethically license."
"Splitting the lineup into reasoning, ranking, and inline lets each model do one thing well," said the Safeguard engineering team. "Griffin reasons about the call graph, Eagle orders the queue, Lion guards the keyboard."
The Griffin family, Eagle, and Lion are all available to existing Safeguard customers today.
About Safeguard
Safeguard is the software supply chain security platform that fuses 11 scanners, a security-only AI model lineup (Griffin · Eagle · Lion), and reachability-aware reasoning to find what pattern scanners miss — from CVEs to candidate zero-days — and to ship the fix with cited reasoning. The platform is built for engineering teams shipping production software and for the regulators auditing them. Learn more at https://safeguard.sh.