AI Security
Griffin AI vs Claude Haiku for Bulk Scanning
Claude Haiku is the cost-efficient model Griffin uses for high-volume scan interpretation. Here's how raw Haiku compares to Haiku inside Griffin's bulk pipeline.
Feb 2, 20266 min read
Deep dives, practical guides, and incident analyses from engineers who build Safeguard. No fluff, no vendor FUD — just what you need to ship secure software.
Claude Haiku is the cost-efficient model Griffin uses for high-volume scan interpretation. Here's how raw Haiku compares to Haiku inside Griffin's bulk pipeline.
DEF CON 33 brought hacker-energy attention to package ecosystems, CI/CD abuse, and AppSec Village. Here is what supply chain defenders should take home.
Researchers tracked a PyPI campaign publishing malicious packages under the mexalz and related account names, targeting Python developers with infostealers.
Dev containers promise reproducibility and isolation. They also pull in a long tail of scripts, dotfiles, and feature repos that most teams never audit. Here is how to fix that.
The metrics that actually distinguish high-functioning application security programs from theater, with concrete formulas and reporting cadences for 2026.
RansomHub absorbed affiliates displaced by BlackCat and ran one of the most prolific extortion operations of 2024. Here is what made its tradecraft effective and how to counter it.
crates.io has gained real supply chain features over the past two years. Here is an honest read on what works, what is still immature, and where to invest.
Safeguard.sh is in early discussions with Sify Technology (USA) to evaluate a joint motion across network services, managed security, and enterprise reach.
The design rationale behind Griffin, Safeguard's triage agent — how the loop is structured, why we bounded reasoning depth, and how tool calls stay auditable.
Weekly insights on software supply chain security, delivered to your inbox.