Remote

Jascha

@jascha@infosec.exchange

mastodon 4.8.0-alpha.3+glitch

Technologist and person of interest. Work in Cybersecurity, AI, and ☁️.

0 Followers

0 Following

10 Posts

Joined June 13, 2025

Github:

https://github.com/jaschadub

BSky:

https://bsky.app/profile/jascha.me

Symbiont:

https://symbiont.dev

ThirdKey:

https://thirdkey.ai

Jascha @jascha@infosec.exchange

· 16mo ago

Hello World! #introduction

Work in cybersec for 25+ years. Big OSS proponent.

Latest projects:

VectorSmuggle is acomprehensive proof-of-concept demonstrating vector-based data exfiltration techniques in AI/ML environments. This project illustrates potential risks in RAG systems and provides tools and concepts for defensive analysis.
https://github.com/jaschadub/VectorSmuggle

SchemaPin protocol for cryptographically signing and verifying AI agent tool schemas to prevent supply-chain attacks (aka MCP Rug Pulls).
https://github.com/ThirdKeyAI/SchemaPin

#ai #AiResearch #aisecurity #rag #mcp #mcpserver

GitHub

GitHub - jaschadub/VectorSmuggle: Research framework that quantifies how steganographic obfuscation of embeddings defeats off-the-shelf statistical detection in RAG pipelines, paired with the VectorPin cryptographic-provenance defense

Research framework that quantifies how steganographic obfuscation of embeddings defeats off-the-shelf statistical detection in RAG pipelines, paired with the VectorPin cryptographic-provenance defe...

Jascha @jascha@infosec.exchange

· 14mo ago

I had this video mysteriously removed from my IG account so just gonna put it here too.

#LAProtests

Jascha @jascha@infosec.exchange

· 14mo ago

After much work I am happy to say that Symbiont Open Source is available.
#ai #agent #agenticai #symbiont #symbi #cybersecurity #cybersec #zerotrust

https://github.com/ThirdKeyAI/Symbiont
https://docs.symbiont.dev

GitHub

GitHub - ThirdKeyAI/Symbiont: Symbiont lets teams build production agentic applications where every action is governed, every tool is verified, every agent has identity, and every decision leaves an audit trail.

Symbiont lets teams build production agentic applications where every action is governed, every tool is verified, every agent has identity, and every decision leaves an audit trail. - ThirdKeyAI/Sy...

Jascha @jascha@infosec.exchange

· 7mo ago

I have been busy building:
https://symbiont.dev
https://agentpin.org
https://schemapin.org

Symbiont

Symbiont — Policy-Governed Agent Runtime

Policy-governed agent runtime for production. Execute AI agents and tools under explicit policy, identity, and audit controls. Written in Rust.

Jascha @jascha@infosec.exchange

· 15mo ago

My most recent post about ever growing catalog of AI related attack vectors with PoC code.

https://research.thirdkey.ai/blog/agentnull-ai-attack-vectors/

ThirdKey’s AgentNull: Unveiling the Growing Catalog of AI Attack Vectors

ThirdKey Research

ThirdKey’s AgentNull: Unveiling the Growing Catalog of AI Attack Vectors

Exploring ThirdKey’s comprehensive catalog of AI attack vectors targeting autonomous agents, RAG pipelines, and vector databases - complete with proof-of-concept demonstrations.

Jascha @jascha@infosec.exchange

· 16mo ago

Hi all show SchemaPin some love on ProductHunt today and use it to help secure your AI Agents!

https://www.producthunt.com/products/schemapin-protocol

producthunt.com

Jascha @jascha@infosec.exchange

· 16mo ago

Did someone say #funkymusic !?!?

https://kissa.fm

kissa.fm

KissaFM Live on Mixcloud

Tune into Kissa.FM's live stream funky vibes and soulful grooves.

Jascha @jascha@infosec.exchange

· 14mo ago

Replying to

@os_sci Can't tell someone to RTFM without the M 👍

Jascha @jascha@infosec.exchange

· 5mo ago

My latest paper:

VectorSmuggle: Steganographic Exfiltration in Embedding Stores and a Cryptographic Provenance Defense

Paper: https://doi.org/10.5281/zenodo.20058255
VectorSmuggle: https://github.com/jaschadub/VectorSmuggle

Zenodo

VectorSmuggle: Steganographic Exfiltration in Embedding Stores and a Cryptographic Provenance Defense

Modern retrieval-augmented generation (RAG) systems convert sensitive content into high-dimensional embeddings and store them in vector databases that treat the resulting numerical artifacts as opaque. Major vector-store products do not appear to provide native or default controls for embedding integrity, ingestion-time distributional anomaly detection, or cryptographic provenance attestation. This paper shows that this opens a class of steganographic exfiltration attacks: an attacker with write