Cybercamp.ai

What happens when AI gets access to your systems, data and decisions?

Cybercamp.ai is a new conference in Brno, part of Dny AI, focused on the intersection of Cybersecurity and AI.

Confirmed talks and workshops

🔐 You Can't Secure What You Don't ControlMarek 'marx' Grác (Principal Research Fellow, Red Hat)This keynote examines how visibility, asset management, access control, AI gateways, and data anonymization enable verifiable AI security and compliance.

🔎 The Scanner Said Clean. We Ran It Again: Stochastic False Negatives in Vulnerability DetectionPetr Simecek (AI Researcher, AISLE)A 25,000-trial study shows that diverse, inexpensive models outperform single models in detecting vulnerabilities, while stateful C bugs remain difficult to rediscover.

🧪 Poison the Well: Real-world context poisoning attacks against AI agentsRadek Hendrych (Staff Security Manager, Productboard)Explores how context poisoning attacks target AI agents, why they are difficult to detect and defend against, and emerging mitigation strategies.

🕵️ Agentic AI in Threat Intelligence: LLMs and BeyondRoman Oravec (Threat Intelligence Specialist, Whalebone)Martin Kuruc (Threat Intelligence Lead, Whalebone)A production case study on using agentic AI, deterministic pipelines, and offline evaluation to detect malicious domains and improve threat intelligence outcomes.

🚧 99% Accuracy, Still Hijacked: Lessons from Building a Custom Prompt Injection GateMarina Volkova (AI Researcher, Emplifi)This talk examines why prompt injection detectors fail in production and presents better datasets, hard negatives, fine-tuning, calibration, and thresholding strategies.

🐳 YOLO mode, isolated by design. How Docker's sandboxes change what "allow all" meansPawel Piwosz (Developer Advocate, UpCloud)A live demonstration of an autonomous AI agent running a dangerous task inside Docker Sandboxes with per-agent microVMs and controlled file, network, and secret boundaries.

📊 The Assurance Gap: A Practical Method for Scoring AI Model TrustJohn Waller (Cyber/AI/Risk Advisory Leader, UltraViolet Cyber)Summary: A six-domain, evidence-weighted method scores AI model assurance, applies overriding gates, and evaluates specific model artifacts rather than model families.

🪪 When AI Agents Become the Attacker: Identity Hijacking Across Autonomous WorkflowsSamet Can Taşcı (Associate Vice President, ISS STOXX)Mehmet Önder Key (Cyber Security Consultant) Examines how attackers can hijack enterprise AI agents through prompt injection and untrusted inputs, exploiting delegated permissions and tool integrations, with defensive mitigation strategies.

📡 The Age of Borg: Assimilating your perimeterCaleb Crable (Principal AI Red Team Engineer, Ethos)A framework uses AI, repurposed phones, and security devices to passively detect, illuminate, and triangulate signal activity around a property at the hardware-address level.

🛡️ Read Everything, Send Nothing: I Gave an AI Agent My LifeTin Marković (Independent software consultant)A talk about Seneschal, an open-source broker that limits AI agents through revocable permissions, human approvals, append-only actions, and protection against prompt injection.

🧰 AI security survival playbook: a practical guideMarek Šottl (CEO of HackiHub)The session presents practical defenses for prompt injection in LLM applications, including NLP input validation, prompt tracing, threat modeling, and layered security controls.

🎯 AI Pentesting: Hype or Reality? Lessons from Building an Autonomous Pentesting SolutionDávid Bolvanský (Software Engineer, Invicti Security)An engineering account of building Invicti’s autonomous AI pentesting solution, examining its capabilities, limitations, architecture, and impact on offensive security.

🌐 Shadow IT 2.0: Kenneth Accidentally Became a VendorJonas Hermansson (Thought Leader, IT-HUSET)How to move employee-built AI applications from experimentation to production without creating unmanaged, business-critical systems.

🕵️ Trust me, Bro: Social Engineering the AI-Agent LoopDanita Samuel Prakash (Product Lead, Cyber Seren)How attackers manipulate human supervisors in AI-agent workflows and why human approval alone is not enough to ensure security.

🛠️ 20 minutes to a more secure agent (Workshop)Radek Hendrych (Staff Security Manager, Productboard)A practical workshop on securing OpenClaw, Hermes, Claude Code, and Codex against accidental data deletion and data leakage while preserving usability.🧠 Surgery on a Brain: Live-Patching Facts in LLMs with ROMEMatej Olexa (Researcher, Brno University of Technology)Explores how ROME enables precise fact editing in LLMs, the associated security risks and unintended effects, and techniques for detecting maliciously modified model weights.

🧩 Closing the Loop: Building a Shared Memory for Your Organisation's AIShaharia Azam (Senior Engineer, Trade Republic)Presents a vendor-neutral, MCP-powered shared memory that allows AI tools to access organisational knowledge and contribute new insights, helping knowledge accumulate across teams.

📄 Ink Only Machines Can Read: Hijacking AI AgentsEmil Khamedov (Quantitative Developer, ATLAS Capital)A controlled experiment and live demo demonstrate how hidden instructions in documents can manipulate AI agents analysing corporate disclosures and generating trading recommendations.

⚔️ Evolution of Offensive AI: From First Prompt to Autonomous HackingPatrik Žák (Ethical Hacker, SYSNETSHIELD)Juraj Danis (Ethical Hacker, SYSNETSHIELD)Traces the evolution of offensive AI, from the early limitations of ChatGPT to autonomous hacking agents capable of gaining enterprise administrator access.

🤖 The Unconstrained Machine: Why AI Keeps Behaving Badly, Exactly as DesignedJohn Waller (Cyber/AI/Risk Advisory Leader, UltraViolet Cyber)The talk presents a four-constraint framework for understanding how AI systems fail through unconstrained optimization, from algorithmic collusion to agentic sandbox escapes.

🧠 How Our Good Instincts Let AI Agents Escape RealityMichal Svoboda, Security & engineering mentorExamines how rigid plans, evidence, and reviews can cause autonomous AI agents to follow scripts instead of adapting to reality.

CybercampAI Cybersecurity AI AISecurity DnyAI Brno