Artificial intelligence is rapidly shifting from simply answering questions to actively executing complex tasks, creating what the tech industry is now calling an agentic workforce. While 85% of major enterprise customers are actively experimenting with these autonomous AI agents, a staggering reality remains: only 5% have actually moved this technology into active production. The primary roadblock preventing mass adoption is a fundamental lack of security and trust. Cisco is stepping in to address this massive operational gap, announcing a comprehensive suite of security solutions designed to safely integrate AI agents into daily corporate workflows.
Establishing Identity and Accountability
To safely deploy an agentic workforce, companies must first establish strict operational boundaries to protect their networks from the agents themselves. Cisco approaches this by extending its Zero Trust Access architecture directly to AI-driven workflows. In a traditional setup, most enterprises lack visibility into which autonomous agents are running on their networks, let alone knowing who is ultimately accountable if an agent makes a critical error.
Through the new capabilities in Cisco Identity Intelligence, organisations can now accurately discover both agentic and non-human identities across their infrastructure. The integration of Duo Identity and Access Management allows companies to register these AI agents and explicitly map them to human owners. This critical step ensures that every digital action is traceable and strictly accountable to a real employee. Furthermore, agents are assigned highly specific permissions for short durations, with all tool traffic securely routed through a model context protocol gateway to eliminate security blind spots.
Hardening Agents Against Manipulation
Beyond restricting what an internal agent can access, organisations must also protect these digital workers from external manipulation and corruption. Traditional scanning tools simply cannot simulate the real-world threats that agents encounter. To counter this, Cisco is expanding its toolset with AI Defense: Explorer Edition, a self-serve platform that democratizes AI safety testing for developers and security researchers. Before deploying an agent into the wild, teams can use this toolkit to red-team their AI models, actively testing their resilience against prompt injections, jailbreaks, and other adversarial attacks.
To further bridge the gap between development and security, Cisco introduced DefenseClaw, an open-source secure agent framework. DefenseClaw automates security scanning and asset inventory directly into the development pipeline, with future plans to integrate with NVIDIA OpenShell to provide robust protection at the runtime level. The company also launched the LLM Security Leaderboard, providing an objective, transparent resource for evaluating how different models handle malicious prompts and manipulation strategies.
Upgrading the Security Operations Centre
As AI-driven threats begin operating at unprecedented speeds, human defenders need advanced AI assistance to keep pace and respond effectively. Splunk, which sits under the Cisco security umbrella, is rolling out a series of innovations designed to transform Security Operations Centres from reactive monitoring stations into proactive defence hubs.
Analysts are often overwhelmed by alert fatigue, but Splunk is mitigating this by introducing specialised AI agents of its own. These include the Detection Builder Agent, Triage Agent, and Guided Response Agent, which actively evaluate situations and execute security workflows rather than simply surfacing raw data for humans to read. Combined with new platform integrations like Exposure Analytics for real-time risk scoring and Detection Studio for streamlining the detection engineering lifecycle, security teams can automate their response workflows and outmaneuver sophisticated adversaries at machine speed. By embedding strict security controls directly into the foundation of these emerging technologies, Cisco aims to give businesses the confidence they need to fully leverage the agentic AI economy without compromising their digital safety.
