Artificial intelligence is moving faster than ever before. Organisations everywhere are rushing to deploy autonomous AI agents to handle daily operations, write code, and streamline customer service. But speed brings new risks. As AI agents grow more powerful, the risk of unexpected vulnerabilities and security breaches grows exponentially. Companies need a reliable way to test systems and maintain control before letting bots loose in production environments.
Cisco announced powerful new capabilities designed to secure what it calls the agentic enterprise at Cisco Live APJ in Melbourne. The updates span Cisco AI Defense, Splunk, and Cisco IQ. Together, these innovations create a continuous trust loop that helps security and operations teams find risks, measure financial value, and prove trust at machine speed. Leaders can finally embrace advanced automation without sacrificing safety or visibility.
Stress-Testing Autonomous Systems with Claude Mythos
One of the most exciting updates is within Cisco AI Defense. The adaptive AI red-teaming framework now includes Anthropic’s Claude Mythos. Red teaming is all about simulating real-world attacks to find weak spots before malicious actors do. By harnessing advanced adversarial reasoning, Claude Mythos helps security teams plan sophisticated tests and translate results into actionable attack ideas for each unique system. This means defenders can stress-test their AI architectures against realistic, multi-step threat scenarios.
Cisco built this integration with strict safeguards in place to ensure comprehensive safety. The system operates exclusively against registered, authorised targets owned by the enterprise customer, preventing any scanning of random internet infrastructure. Every engagement requires authentication and is cryptographically tied to a named operator. Furthermore, sessions are time-bounded and can be halted instantly by human users if needed. This gives organisations the deep testing power of frontier AI models while maintaining total operational oversight.
Tracking Financial Impact with Splunk Tokenomics and Evaluations
Beyond finding risks, businesses also need to know whether their AI investments are actually paying off. Managing runaway costs is a major headache for IT leaders today. To address this, Cisco is introducing enhanced Tokenomics through Splunk Agent Observability. Tokenomics connects AI spending directly to output quality in real time. Using data from Amazon Bedrock and Amazon CloudWatch, organisations can pair each agent’s operating cost with automated evaluations that grade its work. This gives executives a single, unified view of AI return on investment.
To make these evaluations faster and cheaper, Splunk is also rolling out new Eval-G1 evaluation models and an Evaluation Tuner. Teams can turn their agentic traces and expert feedback into fine-tuned evaluator models. These custom models execute guardrails at up to 98% lower cost than traditional commercial language models while delivering response times two and a half times faster. It allows companies to enforce quality standards and steer agents in real time without blowing up their tech budgets.
Connecting Vulnerabilities to Action with Cisco IQ and Data Fabric
Finding a flaw is only half the battle. Security teams must also act quickly to fix underlying infrastructure weaknesses. Cisco is closing the loop by connecting live threat detection data directly with exposed network infrastructure through Cisco IQ and the Cisco Data Fabric powered by the Splunk Platform. As threats compress the window between vulnerability disclosure and exploitation from weeks to hours, static audits are no longer enough.
Through open APIs, Cisco IQ vulnerability and configuration data flows straight into the Splunk platform. Security operations teams can correlate live threat detections with underlying network exposure instantly. Additionally, Model Context Protocol and ServiceNow connectors automatically translate exposure findings into asset-linked problem records. This removes tedious manual data reconciliation and lets IT teams coordinate remediation smoothly across complex corporate networks.
Scaling modern AI adoption requires more than just powerful models. It requires a balanced approach where security, visibility, and financial governance work together seamlessly. By uniting Anthropic’s advanced reasoning models with deep observability and machine-speed infrastructure intelligence, Cisco is giving enterprises the tools they need to move fast without losing control. The future of enterprise AI belongs to organisations that can innovate boldly while keeping trust at the very centre of their operations.
