Blog
What's new in Guardix 0.7
A new audit pipeline. EVMBench recall moved from 59.8% to 88.9%.
Guardix on EVMBench: 88.9% attributed recall
Attributed recall on the completed 40-audit board is 88.9%. That is 104 of 117 high-severity, loss-of-funds bugs. The last public Guardix number was 59.8%.
A self-improving system for security audits
An hours-long audit pipeline that we keep raising on labeled studies: run, score, read the misses, build evals, then change the agents, tools, and environment. The completed EVMBench board is 88.9% attributed recall.
Introducing Release Audits
Check whether the third-party audits behind a launch still cover the code and contracts you're shipping - then share the scorecard.
Model quality update: EVMBench, validation, and fewer vague claims
Guardix publishes quality as benchmarked behavior: 59.8% EVMBench recall, 70 of 117 high-severity bugs detected, and a clear plan for the misses.
Guardix on EVMBench: 59.8% recall across 117 high-severity vulnerabilities
We ran the full Guardix pipeline against EVMBench — 117 ground-truth vulnerabilities across 40 Code4rena audits — with zero modifications. Here are the results, what we learned, and what we are improving.
Inside the exploit pipeline: how Guardix proves vulnerabilities are real
Finding a vulnerability is one thing. Proving it is exploitable on a live fork with real protocol state is another. Here is how the exploit pipeline works end to end.
Ship, audit, fix, repeat: the continuous security workflow
How versioned audit runs and immutable scan history turn security review into a continuous practice instead of a one-time gate.
Architecture-first auditing
Why understanding invariants, assumptions, and design decisions before searching for bugs produces better results.
The economics of smart contract security
Traditional audits cost $50K+ and take weeks. AI-first auditing changes the math for teams that ship continuously.
Consensus validation: how multi-model agreement reduces noise
A single model produces noise. Multiple models with independent analysis and cross-validation produce signal.
Introducing Guardix
Why we built a repository-native audit platform and what changes when security review starts from your actual codebase.
No posts in this category yet. View all posts