Agentic Security Series
Understanding the risks of autonomous AI systems
As AI agents gain autonomy—accessing files, sending messages, executing code—their attack surface expands dramatically. This series explores real-world threats, industry standards, and the architectural patterns that make agents safer.
The Series
The Moltbot Moment: Why Persistent Memory Changes Everything
How a viral open-source agent revealed the fourth risk factor in agentic AI security.
The OWASP Agentic Top 10: A Developer's Field Guide
A practical walkthrough of the 2025 standard for agentic application security.
Prompt Injection in the Wild: Real Attack Patterns
How attackers exploit the gap between instructions and data in LLM-powered systems.
The Case for Privilege Separation in AI Agents
Why the principle of least privilege matters more than ever in agentic systems.
Git-Verified Agents: Closing the Supply Chain Gap
Using version control and cryptographic verification to trust what your agents run.
Building Auditable Agents: Receipts, Rankings, and Runtime Monitoring
Cryptographic audit trails and continuous monitoring for production agent systems.
Supply Chain Vulnerabilities: Poisoned Tools and Malicious MCP Servers
How attackers compromise the agent supply chain through malicious tool descriptors, compromised templates, and dependency confusion.
Unexpected Code Execution: Eval, Sandbox Escapes, and SSTI
When agents generate and execute code dynamically, every output becomes a potential attack vector.
Insecure Inter-Agent Communication: Spoofing, Tampering, and Lateral Movement
When agents talk to agents, every message is an attack surface. Spoofed identities, tampered payloads, and unauthorized delegation.
Cascading Failures: When One Agent's Problem Becomes Everyone's Crisis
Retry storms, error propagation, and resource exhaustion in multi-agent systems.
New posts published regularly. Check back for updates.