Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Blueberry Muffin: Agent Memory Injection
Discover how attackers can inject lies into AI agent memory, causing hallucinations, weaponizing trust, and triggering logic denial-of-service attacks, leading to system paralysis.
AI agents currently cannot tell the difference between their own memories and malicious lies planted by attackers. By simply appending a single line of text to an agent’s memory file, I was able to:
- 𝗣𝗼𝗶𝘀𝗼𝗻 𝗶𝘁𝘀 𝗿𝗲𝗮𝗹𝗶𝘁𝘆: Force a production-grade agent to hallucinate technical definitions.
- 𝗪𝗲𝗮𝗽𝗼𝗻𝗶𝘇𝗲 𝘁𝗿𝘂𝘀𝘁: The agent cited my own research back to me to “sell” the lie.
- 𝗧𝗿𝗶𝗴𝗴𝗲𝗿 𝗮 𝗟𝗼𝗴𝗶𝗰 𝗗𝗼𝗦: Trap it in an infinite loop until it paralyzed itself.
𝗕𝗿𝗶𝗹𝗹𝗶𝗮𝗻𝘁 𝗔𝗺𝗻𝗲𝘀𝗶𝗮 - how the agent ‘engineered’ an OOM and then proceeded to delete my codespace.
- AntigravityGoogle Antigravity is an agentic development platform that automates the software lifecycle: from autonomous planning and compilation to multi-cloud deployment.Antigravity transforms the IDE into a mission control for autonomous agents: handling everything from initial scaffolding to cloud-ready deployment. You define the mission (e.g., 'build and containerize a FastAPI backend') and the system executes the plan across your local terminal and browser. It manages the messy details: resolving compilation conflicts, configuring Dockerfiles, and pushing to platforms like Firebase or Google Cloud. You verify progress through Artifacts (task lists and live walkthroughs) that ensure the agent’s logic aligns with your requirements. It’s about moving fast: leveraging Gemini 3 and Claude 4.5 to turn high-level prompts into production-grade artifacts without the standard setup friction.
Related projects
DeepCache - An Agentic Framework for Neural Compression
Toronto
The talk explains how to attach complexity metadata to agent messages, use it for compression, verify integrity, predict…
Human-ish: A Mirror for the AI Age
Toronto
This talk introduces Human-ish, a tool to detect AI-written LinkedIn posts, exploring the feedback loop between AI content…
Beyond Text: Hacking Transformers to Detect Anomalies in Million-Scale Netflow Data
Toronto
See how ELECTRA was adapted to classify million-scale netflow data by treating traffic as text, graphs, and quantum…
Context Engineering for Stateful AI Agents
Toronto
Learn context engineering to build stateful AI agents using Sessions for short-term history and Memory for long-term personalization…
Observability that drives true ROI
Toronto
This talk introduces AI managers that monitor and summarize agent behavior, moving beyond dashboards to quickly show what…
Multi-Agent Fraud Detection: When LLMs Argue About Bitcoin Laundering
Toronto
Three LLM agents debate Bitcoin fraud using Graph RAG on Neo4j, presenting prosecutor, defense, and judge roles for…