BASTYN: Build AI products your users can trust

I spent 20+ years building regulated consumer products inside Fortune 100 companies. Then I started vibe coding, joined hackathons, and was shocked at how easily teams shipped AI agents with zero independent safety testing. Just great vibes! I went looking for the standard. In 2024, they didn't exist. So we built them.... We aggregated all major standards and frameworks( EU AI Act, NIST, ISO, OWASP, MITRE etc) to define what "safe" actually looks like for the new threat surface exposed by agentic systems. We partnered with University academics and this has now evolved into 23 test vectors, 717 adversarial probes as we kept finding critical vulnerabilities in products that were deemed to be 100% safe (not through their code but the poisoning from tools, interactions and dependencies that evolve over time). We then built a risk grading algorithm that considers the functional context, jurisdiction, and domain of the AI system so we can help teams meeting their compliance needs with a one-click evidence pack for conformity assessment while protecting against the attack surface they cannot see. We test what it does when someone actively uses it or tries to break it. The thing that surprised me most on the journey is AI doesn't just fail passively, it can actively find new ways around guardrails after you ship. BASTYN runs in your CI/CD pipeline before you merge and acts as an SSL certificate for AI trust when live. Always on. Publicly verifiable. Revoked automatically if your agent drifts. BASTYN covers multiple use cases: - you are building AI Agents and want verification of safety before users become your test suite - you are using AI features in your product - you have shipped AI features and not sure what they've been doing since launch - you are evaluating a third-party AI agent to embed/ use in your organisation and you want to prove that it is safe before use. Chat to me directly Bal@Rootedlogic.ai