AI + Cyber: Breaking the Myth(os)
Cybersecurity and GenAI are colliding — and this is where it breaks.
The same frontier models that surface thousands of zero-days get weaponized by a single malicious skill file. The same agents you're shipping to production are the ones attackers are learning to hijack.
Three practitioners, three live demos of things actually breaking, one evening on Floor 28.
🍕 Food, drinks, and networking included. 🙏 Thanks to AWS for hosting and sponsoring the venue (GenAI Israel is the organizer — this is a community event, not an AWS event).
🗓️ Agenda
18:00–18:45 — Networking & food
18:45–18:50 — Welcome
18:50–21:00 — Talks
🎤 Talks
Breaking the Myth(o)s
Aviad Chen — Co-Founder & CTO @ Backline
Claude Mythos is Anthropic's newest frontier model — good enough at understanding and modifying software to have surfaced thousands of zero-days, and for the US government to suspend access days after release. Project Glasswing puts that capability in defenders' hands before attackers get the same edge.
Aviad covers what Mythos actually is, what Glasswing is trying to achieve, and real examples of it finding exploitable bugs across vulnerability types. Then the part the demos skip: finding is the easy half. A patch that's correct in isolation breaks your build. A misconfig fix locks out a service. Tightening one permission kills an integration two teams away. Why a single model patching in a vacuum doesn't scale — and what real remediation takes.
Co-Founder & CTO at Backline AI. Mamram grad, career spent at the intersection of engineering and cyber, including key roles at two acquired startups — Whitebox Security and Apolicy. Now focused on how AI reshapes defensive security and autonomous exposure remediation.
Securing GenAI in Production: OWASP Risks Meet AWS Guardrails
Eitan Sela — Principal GenAI/ML Specialist Solutions Architect @ AWS
The OWASP Top 10 for LLM Applications, and how to defend against it on AWS. Live demo of Amazon Bedrock Guardrails blocking harmful inputs and outputs, plus how AgentCore Runtime Isolation and AgentCore Policy enforce security boundaries for autonomous agents. You'll leave with best practices for securing GenAI workloads in production.
Principal GenAI/ML Specialist Solutions Architect at AWS, helping customers build and operate GenAI and ML on AWS. Avid marathon runner.
Your AI Agent Installed Malware Because a SKILL.md Told It To
Liran Tal — AI Security Researcher @ Snyk
That SKILL.md you just installed to supercharge your coding agent might be exfiltrating your AWS credentials right now. Just like early npm, attackers are abusing Agent Skill ecosystems for malware campaigns — and ToxicSkills research found 37% of nearly 4,000 skills carrying malware or security weaknesses, including one "security scanner" that was itself malware.
Liran runs live hacking: a malicious skill fooling a coding agent into rogue actions, a prompt injection leaking secrets over email, a leaky skill passing credit card numbers straight through the LLM context. Then he flips to defense — detecting dangerous SKILL.md files and catching what every regex scanner misses. You'll leave with a threat model for Agent Skills supply chains and the tools to audit your own agents.
AI Security researcher and Head of Developer Relations at Snyk. Worked on the ToxicSkills research, publishes practices for securing MCP servers, and reports CVEs across AI frameworks. Longtime Node.js developer, GitHub Star, and OpenJS "Pathfinder for Security."