AI engineering
Build things, inspect how they work, then pull the thread.
Shmulik CohenAI × SECURITY × OPEN SOURCE
AI engineer · Security researcher
Shmulik CohenBuild.
Break.
Understand.
I’m Shmulik Cohen. I build AI tools, investigate software vulnerabilities, and write about the experiments.
AI engineer at Apiiro · Author of AI SuperheroExplore the Agent Skills Network ↗Three paths through my work
Start here.
Three recommended stops for each interest.
Security research
Trace real boundaries and read the evidence.
Experiments
Model steering, a brain simulation, and computational mathematics.
01 / Selected projects & research
Selected work.
Argo CD MCP: unauthenticated cluster access
An investigation into the authentication boundary between an AI tool and the infrastructure it controls.
EnglishSecurity practitioners · Builders
Security / MCPTake a look
Hebrew Codenames
A semantic-alignment experiment using Hebrew word embeddings, model comparisons, and an interactive map.
EnglishBuilders
AI / EmbeddingsTake a look
Captain Obvious
A deterministic scanner for tests that can never fail, using type-checker and program-structure analysis.
EnglishBuilders
Open source / TestingTake a look
Jewish Qwen
Experiments in steering Qwen3.5-4B with neurons, activation directions, and optimized vectors.
EnglishAI-curious · Builders
AI / InterpretabilityTake a look
The repositories changed. The malware fleet stayed.
Following README link changes across the FakeGit software-supply-chain campaign.
EnglishSecurity practitioners · Builders
Security / Supply chainTake a look
Agent Skills Network
An interactive 3D map of agent skills, grouped by semantic similarity.
EnglishBuilders
AI / VisualizationTake a look
LLM sampling
How an LLM picks its next word.
Read the article ↗LLM sampling
How an LLM picks its next word.
Prompt
I asked my AI agent to fix a bug. It decided to…
Sharper at low values; wider at high values.
Keep up to K candidates.
Keep the likely mass.
Drop low relative scores.
This demo: temperature → Top-K → Top-P → Min-P.
Illustrative scores, not a live model.
Listen to the Hebrew episode ↗Choose a scenario, then sample.
5 of 5 candidates remain
debug
refactor
document
panic
dance
Speaking
Talks & podcasts.
מעבר למודל: מי בוחר את המילה הבאה?
Guest conversation with Dekel Asaf and Shiron Babi about sampling, softmax, temperature, Top-K, Top-P, and Min-P. Episode 4, in Hebrew.
Hebrew onlyBuilders · AI-curious
AI / SamplingOpen original
Codex Pets: playful agents, real possibilities
My talk at the first Codex Community Meetup at Wix, covering Codex Pets, Guardian, and the possibilities opened by a shared protocol.
Hebrew onlyBuilders · AI-curious
AI / ExperimentsTake a look
פרומפט וחצי
My Hebrew podcast about AI. Listen to the show on Spotify.
Hebrew onlyAI-curious · Builders
AI / ConversationOpen original
02 / Writing
Articles & essays.
010203
הצייד והמכונה: גילוי חולשות ו-Bug Bounty בעזרת AI Agents
Evil Qwen’s Jewish Cousins
I Don’t Want to Become a Meat Proxy
AI Superhero
Read on Substack ↗My newsletter on AI, software, and security.
About
AI engineering meets
security research.
I’m an AI engineer at Apiiro, a security researcher, an open-source builder, and the person behind AI Superhero.
More about me ↗