Built by a Security Practitioner.
For Security Teams.
DefendML is founder-led, with human accountability for how tests are designed, evidence is interpreted, and releases are approved.
AI-assisted internal workflows are tools, not fictional employees or independent operators.
Sareth Dustin Sovan
Founder & Chief Architect
20 years in IT operations and security • 4 SOC 2 audits
Leads DefendML product architecture and remains accountable for the testing methodology, evidence model, and release decisions. AI-assisted internal workflows support the work; human review owns the outcome.
DefendML uses AI-assisted workflows for research, analysis, and implementation. The founder remains accountable for product decisions, customer claims, evidence, and every production release.
Interested in working with us? Get in touch →How We Work
Four principles that drive every product and engineering decision at DefendML.
Attack First
We find vulnerabilities before adversaries do. Every product decision starts with the attacker's perspective.
Evidence Over Assertions
Claims that an AI system is safe are not enough. We produce structured, auditable evidence that security teams can stand behind.
Honest Positioning
We test AI applications pre-deployment. We don't provide runtime protection, and we don't overstate what a single scan can prove.
Practitioner-Built
DefendML is founder-led, grounded in 20 years of IT operations and security experience, including four SOC 2 audits.
Join Our Mission
We're building the offensive AI red team testing layer the security industry needs — starting with a library of 415 adversarial scenarios and evidence reports that hold up to enterprise scrutiny.
DefendML is a founder-led service. We welcome conversations with security engineers and AI practitioners who want to improve evidence-backed offensive testing. If that's you, reach out.