Journal

Records from the dark.

Analyses on the security of AI systems: prompt injection, data poisoning, secure AI architecture and regulation. English editions of the German Journal.

2026

J-009 / 2026-09

Navier-Stokes solved? What OpenAI's proof shows, and what it does not

Regulation & Governance J-008 / 2026-09

LLM Jailbreaks: six technique classes, one test plan

Jailbreaks J-007 / 2026-09

LLM Red Teaming: what a test proves and what only reassures

Red Teaming J-006 / 2026-09

OWASP LLM Top 10: all ten risks explained, 2026 edition

Secure AI Architecture J-005 / 2026-09

LLM Guardrails: 95% protection, 100% evasion, both measured

Guardrails J-004 / 2026-09

The OpenAI-Hugging Face incident: how the AI agents reasoned

Secure AI Architecture J-003 / 2026-08

MCP Security: how a single server hijacks the agent

AI Agents J-002 / 2026-06

Indirect Prompt Injection: when foreign content steers the AI

Prompt Injection J-001 / 2026-05

AI agents as a security risk: the attack surface

AI Agents

> more records in progress … this journal keeps growing.