Cisco announced two major security initiatives at the RSA Conference 2026: DefenseClaw, a comprehensive framework for securing AI agents in production environments, and Zero Trust Access, an architecture that applies zero trust principles specifically to agentic AI systems. The announcements address an emerging security crisis in enterprise AI: while 85% of enterprises are experimenting with AI agents, fewer than 5% have deployed them in production at meaningful scale. The gap between experimentation and production reflects a critical missing piece in the enterprise AI stack: security frameworks designed from the ground up for systems where AI agents make autonomous decisions and take autonomous actions.
DefenseClaw represents Cisco's answer to the question of how to secure delegated AI agent autonomy. Unlike traditional security frameworks designed for human-initiated requests, DefenseClaw must contend with agents making decisions in milliseconds, operating across multiple cloud platforms simultaneously, and potentially spawning secondary agents to complete delegated tasks. The framework consists of three layers: identity and access control for agent interactions, behavioral analysis to detect anomalous agent actions, and sandboxed execution environments where agent operations can be monitored and revoked in real time. The name itself—DefenseClaw—evokes controlled autonomy, a claw that operates with precision under constrained rules.
Zero Trust Access extends the zero trust security model—which assumes all network access is untrusted until verified—to AI agent interactions. In traditional zero trust implementations, humans authenticate and are granted access based on their identity and context. In the agentic version, each agent action (whether reading a database, initiating a payment, or modifying a configuration) is treated as an untrusted request requiring real-time verification against defined policies. Cisco's implementation includes continuous verification of agent behavior, dynamic policy enforcement based on context, and audit trails that record not just what actions occurred but why the agent took them and what constraints were applied.
The timing of Cisco's announcement is strategically significant. Enterprise CISOs are increasingly anxious about AI agent deployment, particularly as the business case for agents becomes clearer. Cost savings from automating routine tasks—from customer support to procurement to financial reconciliation—can reach 30-40%. But deploying these systems without robust security frameworks creates enormous liability. A compromised AI agent could inadvertently approve fraudulent transactions, share confidential data with unintended recipients, or disable critical infrastructure. Cisco's frameworks are explicitly designed to make these failure modes detectable and remediable in real time.
What makes DefenseClaw and Zero Trust Access potentially transformative is their recognition that AI agent security is fundamentally different from traditional application security. Agents operate with bounded autonomy; they can make decisions within defined parameters but shouldn't be able to exceed them. They also operate under opacity; even their creators can't always explain exactly why an agent made a specific decision in a specific context. Cisco's frameworks embed explainability and auditability into the security architecture, ensuring that even if an agent's decision-making process is opaque, the security guardrails around that decision-making remain transparent and enforceable.
Enterprise adoption of DefenseClaw and Zero Trust Access will likely accelerate throughout 2026 and 2027. Financial services companies, in particular, face regulatory pressure to demonstrate that deployed AI agents operate within defined risk parameters. Healthcare organizations deploying agents for billing and claims management face similar scrutiny from compliance and legal teams. Manufacturing companies deploying agents for supply chain optimization need to ensure that agent decisions don't violate contractual obligations or create unintended operational risks. Cisco's frameworks provide the technical foundation that allows these organizations to move from 5% production deployment of agents to 50%+ with measurable confidence in security and governance.
The broader implication is that enterprise security infrastructure—not just application-level features—is becoming a prerequisite for AI agent adoption at scale. Security frameworks developed for static applications and predictable workflows prove inadequate for autonomous systems operating under uncertainty. Companies that can successfully embed zero trust principles and continuous behavioral analysis into their AI agent infrastructure will gain competitive advantage not just in security posture but in the speed and confidence with which they can deploy new agent capabilities.