DSOLG September 2026 - 9th Anniversary event
About the event
Details
Welcome to the DevSecOps London Gathering March Event on Wednesday 16 September. Along with our amazing talks we bring you the usual conversations, pizza and beer! And because this is our anniversary event, there may be cake!!
Location:
• 📍 Venue: Civo Tech Junction, 1st Floor, 32-37 Cowper Street, London, EC2A 4AW • 📅 Date: Wednesday, 16 September 2026 • 🕕 Time: 6:00 PM – 8:00 PM
Talk 1 - Darshan Patil - Poisoning the training ground
Abstract: As more teams shift toward AI-native workflows, we're adding ML models to the delivery pipeline. Label flipping attacks are a silent data supply-chain threat—they corrupt training labels, degrade model performance, and traditional AppSec tooling misses them entirely.
I want to walk the audience through: • What label flipping looks like (attack demo in PyTorch) • Discuss Defenses • Detecting label flipping\, baking in defense etc.
It's hands-on, practical, and designed around "what can I implement Monday morning"—not pure theory.
I'm a practitioner, not a vendor—just sharing what we're learning in the trenches at a fintech company dealing with these exact concerns. Speaker Bio: Darshan is an application security engineer with a software development background, specialising in integrating security throughout the SDLC. Experienced in SAST, DAST, IAST, and SCA tools including Burp Suite, OWASP ZAP, Semgrep, GitHub Advanced Security, and SonarQube. Skilled in threat modeling, secure code reviews, penetration testing, API security, DevSecOps, and cloud security across AWS, Azure, and GCP. Focused on security automation with Python and Go, CI/CD security, and developer enablement. Experienced in securing AI applications on AWS Bedrock, implementing guardrails, and testing for prompt injection and jailbreak vulnerabilities aligned with OWASP LLM guidance. Passionate about mentoring, security advocacy, and driving impactful security initiatives..
Talk 2 - Sunny Chau - The Agent's inner monologue
Abstract Most agent defenses sit at the edges - input filters, output classifiers, hardened system prompts. Nobody is watching what the agent is actually thinking. And the attack surface that creates is real: indirect prompt injection, tool poisoning, reasoning trace manipulation, exfiltration dressed as legitimate tool use. Recent research from OpenAI and DeepMind found that monitoring an agent's chain of thought improves attack detection by 35% and is 4x more likely to catch nuanced compromise than action-level monitoring alone. Drawing on seven years in offensive security and red team leadership, I'll walk through live attacks against production agents and show how Adrian, Secure Agentics' open-source runtime monitor, catches and stops each one by watching the inner monologue — not just the output. Speaker Bio: Sunny has 4 years of offensive cyber security experience working as a penetration tester and a covert red teamer, having spoken at various security conferences around the world including BSides London, Athens, Hong Kong, Beac0n and more on cloud-native security, identity hacking and all that good stuff. He took his adversarial mind to Secure Agentics as Head of Research, where he leads work on the agentic attack surface, finding the ways autonomous agents get compromised in production, and building the detection logic to catch it.
