CrewAI Security: Protect Multi-Agent Workflows
Two-layer security for multi-agent CrewAI workflows
CrewAI orchestrates multiple AI agents working together — each with their own role, tools, and objectives. But multi-agent systems multiply the attack surface. An injection targeting Agent A can propagate through the crew as Agent B trusts Agent A's output. Rune wraps your entire crew with two layers of protection: crew-level I/O scanning and individual tool call interception for every agent.
Add Security in Minutes
from rune import Shield
from rune.integrations.crewai import shield_crew
shield = Shield(api_key="rune_live_xxx")
protected_crew = shield_crew(my_crew, shield=shield)
# Entire crew is now protected — all agents, all tools
result = protected_crew.kickoff(inputs={"question": "..."})Full setup guide in the documentation
Why CrewAI Agents Need Runtime Security
Multi-agent systems are uniquely vulnerable to cascading attacks. When Agent A passes output to Agent B, any injected instructions in Agent A's output become trusted input for Agent B. A single compromised agent can hijack the entire crew's behavior — and the more agents in the crew, the larger the blast radius.
Top Threats to CrewAI Agents
A compromised agent passes manipulated output to the next agent in the crew. The receiving agent treats it as trusted input, propagating the attack through the entire workflow.
Attackers exploit the interaction between multiple agents' tools. Agent A reads sensitive data, passes it to Agent B, which sends it to an external API — each step looks benign individually.
Injection causes an agent to act outside its defined role — a 'researcher' agent starts executing code, or a 'writer' agent starts making API calls it shouldn't have access to.
What Rune Does for CrewAI
Two-Layer Interception
Rune scans at both the crew level (inputs/outputs of the entire workflow) and the tool level (individual tool calls by each agent). Nothing slips through.
Per-Agent Tool Wrapping
Every agent's tools are individually wrapped for scanning. Each agent gets its own security context and policies — the researcher can read but not write, the writer can draft but not deploy.
Crew Workflow Monitoring
Tracks the flow of data between agents. Detects when output from one agent contains content that could manipulate the next agent in the chain.
Async Crew Support
Full support for both synchronous kickoff() and async kickoff_async() execution. Security scanning adds minimal overhead to crew execution time.
Common CrewAI Use Cases
- Multi-agent research and analysis workflows
- Autonomous coding crews with review and deployment agents
- Content creation pipelines with research, writing, and editing agents
- Business process automation with specialized agent roles
Other Integrations
Secure your CrewAI agents today
Add runtime security to your CrewAI agents in under 5 minutes. Free tier includes 10,000 events per month.