Security research, technical deep-dives, and insights on AI agent security from the Inkog team.
Regulation (EU) 2023/1230 applies January 20, 2027 and pulls AI safety components into a third-party conformity-assessment regime. What it demands in code, who falls under it, and how the rules overlap with the EU AI Act.
What EU AI Act Article 14 actually demands in agent code: a deep-dive on the four oversight requirements, with passing and failing patterns side by side.
EU AI Act Article 15 in agent code: input validation, output validation, error handling, and resource limits — failing and passing patterns teams need.
A reproducible proof-of-concept: a tampered AGENTS.md and a malicious MCP server combine into a zero-click session hijack that exfiltrates .env secrets through an AI coding agent. Why classical SAST misses it, and how declares-vs-does verification stops it.
A compromised AI tool called Context.ai gave attackers access to Vercel customer credentials through a single OAuth grant. Full attack chain, timeline, and response checklists for developers, security teams, and CISOs.
We scanned 561 open-source AI agent repositories, drove the false-positive rate to zero, and opened a disclosure pipeline. Here is what we learned — methodology, top patterns, and the raw numbers.
A walkthrough of the dev-flow security loop: build an agent in Claude Code, scan it with the Inkog MCP, explain and fix findings in the same conversation.
We analyzed CrewAI's delegation system and found unsigned agent-to-agent communication. Here's why multi-agent workflows need message authentication.
We scanned 500+ open-source AI agent repositories for security vulnerabilities. The results reveal a systemic gap between AI agent adoption and AI agent security.
OpenAI acquired Promptfoo. What changes for teams that need vendor-independent AI agent security, and how static analysis fills the gaps eval frameworks leave open.
Inkog Deep goes beyond pattern matching — it understands your agent's purpose, maps its architecture, and explains why findings matter.
The EU AI Act enforcement begins August 2, 2026. A practical checklist covering Article 14 (Human Oversight), Article 15 (Robustness), risk classification, and automated compliance monitoring with GitHub Actions.
Claude, Copilot, and Cursor are great at reviewing code — but they are not security scanners. Six gaps between AI code review and automated security scanning for AI agents.
NIST announced the AI Agent Standards Initiative on February 18, 2026 — the first US federal effort targeting AI agent governance. Key deadlines, pillars, and how to prepare.
A practical guide to detecting and preventing prompt injection attacks in LLM-powered applications, with code examples and YAML rules.
Find the vulnerabilities we write about — in your own code.
15 min · Live Deep Scan on your code · Walkthrough of the findings