News and articles from the Metalware team.
TL;DR: A 13M-parameter model trained from scratch recovers global objects in stripped firmware more accurately than GPT-6.1-Sol (81.8 vs 69.3 Bounds+Kind F1 on 53 real-world images), over 100x faster and runnable locally.
Andrew Nedea
CTO
MMIO fingerprinting with SVDs for memory map prediction and augmenting results with LLMs
Samuel Lichtenheld
Fuzzing firmware is hard. The binaries are stripped, the architectures exotic and there’s often no OS abstractions to lean on.
Andrew Nedea
CTO
Metalware has been named one of the Top 10 Finalists for the 20th annual RSAC 2025 Conference Innovation Sandbox contest.
Ryan Chow
CEO