Cisco launches DefenseClaw and Zero Trust security to protect the agentic AI workforce

Cisco Tackles the Agentic AI Security Crisis

Cisco announced a comprehensive suite of AI security innovations at the RSA Conference 2026 in San Francisco, directly addressing why only 5% of enterprise agentic AI has transitioned from testing to production 1

1

. The networking and security giant introduced DefenseClaw, an open-source secure agent framework, alongside Zero Trust for AI agents and enhanced capabilities across its security portfolio 2

2

. According to DJ Sampath, Cisco's senior vice president of AI and software platform business, DefenseClaw serves as the "operational layer" for agentic security that enables organizations to move from "zero to governed claw in under five minutes" 1

1

.

DefenseClaw: Three-Pronged Protection Against Vulnerabilities in AI Agents

DefenseClaw performs three critical security functions to protect against vulnerabilities in AI agents. First, it scans every piece of code before execution, including every skill, tool, plugin, and code generated by the agent using tools like Cisco's open-source skill-scanner 1

1

. Second, the framework detects cybersecurity risks by scanning all messages entering and leaving the agent at runtime 1

1

. Third, DefenseClaw automatically blocks risky operations, revoking sandbox permissions and quarantining files within 2 seconds without requiring agent restarts 2

2

. The tool integrates with Nvidia's OpenShell sandbox and sends telemetry to Splunk through a pre-packaged connector, ensuring end-to-end security across AI actions 2

2

.

Zero Trust for AI Agents: From Access Control to Action Control

Cisco extended its Zero Trust Access capabilities to the agentic AI workforce, representing what Tom Gillis, senior vice president and general manager for Cisco's infrastructure and security group, calls a "big step forward" for the industry 3

3

. The fundamental shift moves security from access control to action control, providing task-based permissions for agentic activities rather than long-lived credentials 3

3

. Organizations can now register AI agents in Duo IAM and map them to accountable human workers, while Cisco Identity Intelligence discovers agentic identities and related security issues 3

3

. The platform enforces strict policy enforcement through MCP servers, enabling administrators to define rules specifying which tools agents may access and how they may interact with each tool 2

2

.

AI Defense: Explorer Edition and LLM Security Leaderboard

Cisco introduced AI Defense: Explorer Edition, a free version designed to help developers test model and application resilience against jailbreaks and adversarial attacks before deployment 3

3

. The tool includes dynamic red teaming for agents, model and application security testing, and streamlined security reporting 3

3

. Partners can use this discovery tool to demonstrate vulnerabilities in AI agents to customers, creating opportunities for engagement 3

3

. Additionally, Cisco launched the LLM Security Leaderboard, which evaluates model risk and susceptibility to adversarial attacks by contextualizing performance metrics against how models handle malicious prompts and manipulation strategies 4

4

.

Splunk Integration and Machine-Speed Response

Cisco leverages Splunk as the monitoring system of record for all agents, with every agent "born observable" and streaming structured events into Splunk the moment it comes online 1

1

. The company announced a Guided Response Agent, due in alpha release soon, designed to help security operations center teams move from detection hypothesis to production in minutes 1

1

. New Splunk AI innovations transform security operations by automating response workflows and enabling teams to detect and respond to AI incidents at machine speed 5

5

. The platform can now generate an inventory of all network assets complete with data on how they interact with one another 2

2

.

Why This Matters for Enterprise AI Adoption

The security innovations directly address the critical adoption barrier facing enterprises experimenting with AI agents. In Cisco's recent survey of major enterprise customers, 85% reported experimenting with AI agents, yet only 5% had moved agentic technology into production 5

5

. According to the 2025 Cisco Talos Year in Review, attackers increasingly target components that authenticate users and enforce access decisions, a trend expected to accelerate with agentic workloads 5

5

. DefenseClaw will be available on GitHub starting March 27, enabling rapid deployment of the open-source secure agent framework 1

1

. Organizations should monitor how code scanning and Identity and Access Management (IAM) capabilities evolve to handle the unique challenges of securing autonomous agents that act rather than simply respond to queries.