Agent Security

Language models make security decisions about tools that firewalls and code scans cannot detect.

Agents running unsupervised need security controls humans never did.

How Anthropic embedded reasoning into AI safety rules instead of just listing constraints.

A breakdown of five threat domains where autonomous AI agents lose control.

Malicious tool descriptions let AI agents execute attacker commands with production access.

Agentic AI systems break NIST's risk framework at every function.

Autonomous AI agents face a qualitatively different threat landscape than single-turn models.

Why thousands of exposed MCP servers put your data at risk.

Attackers exploit AI agent identities through tool poisoning and prompt injection at runtime.

A working methodology for testing tool-calling agents against four distinct threat classes.

Four threat classes—prompt injection, tool misuse, credential exposure.

Enterprises must govern AI agents before shadow deployments outpace security controls.

Attackers exploit MCP's trusted tool outputs to inject hidden commands into agent reasoning.

Anthropic's integration protocol scales fast, but security wasn't designed in.

AI agents quietly expand their own permissions through incremental steps that each seem reasonable.

Tool calls execute with real credentials, creating exfiltration risks keyword filters cannot detect.

Agents gradually pursue different goals as memory and context reshape their objectives over time.

A new OWASP framework identifies ten critical security risks specific to AI agent tool protocols.

Attackers can hijack agents through persistent injection across tool calls and data pipelines.

Agent workflows bypass traditional perimeter controls and need policy embedded in every tool call.

Attackers embed malicious instructions in tool descriptions to hijack AI agent behavior undetected.