Cisco Report: Trust, Not Speed, Decides Whether AI Agents Get Used

AI agents now act on their own around the clock, and that has shifted the question for business leaders from how fast they can move to how they stay in control, according to a new Cisco report released at Cisco Live Melbourne.

The study draws on interviews with 25 Australian leaders. Sarah Carney, Head of AI at AustralianSuper, told Cisco: "If you haven't built trust, you're not going to get the outcomes or benefits you predicted, because people won't use the system." Cisco frames the lesson simply: capability decides what a business can automate, and trust decides what it will.

Cisco's answer rests on an idea it says has guided it for more than 40 years: the network matters more than any single node. Models, apps and agents all count, but the connections between them are where control lives. Every time an agent acts, it makes a connection, crosses a security boundary and leaves a data trail. When networking, security and visibility work together, Cisco argues, businesses can see what agents are doing and step in at the same speed the AI operates.

The company's announcements address two questions leaders ask before handing AI serious work: what happens if someone breaks in, and can costs be kept under control. On security, Cisco AI Defense is adding Anthropic's Claude Mythos to its red teaming framework, which tests systems for weak spots by simulating attacks. Cisco says the tool is restricted to systems a customer owns and has authorized, that each test is tied to a named person and can be stopped at any time, and that customers cannot prompt it freely. It is coming soon to select customers.

Cisco IQ, which the company says more than 16,000 customers use to monitor close to 20 million devices, now draws on Cisco Data Fabric powered by the Splunk Platform. That lets security teams match a live threat in Splunk Enterprise Security to the device at risk. Connectors for supported AI agents and ServiceNow turn findings into problem records linked to the affected equipment, and an AI-powered playbook flags risks across the equipment's life, from known defects to quantum readiness.

On cost, Splunk Agent Observability gives every agent a receipt and a report card. Tokenomics, a feature Cisco says uses data from Amazon Bedrock and Amazon CloudWatch, puts what each agent costs next to how well it performed. Customers can also turn agent activity and expert feedback into their own evaluator models. Cisco says its next-generation Eval-G1 models run checks and guardrails at up to 98% lower cost than a commercial large language model and respond 2.5 times faster. The evaluation models and tuner are available this month, with Tokenomics set for general availability in early 2027 on AWS Marketplace.

Across Asia Pacific, Japan and Greater China, Cisco notes, markets have taken different paths to modernization, some skipping generations and others upgrading mature networks. AI reaches all those starting points at once while dozens of markets write their own data and AI rules. The company's conclusion: nobody hands over permission to scale. It is earned each time a business tests its AI, fixes what it finds and proves the investment is worth it.