Cisco Adds Claude Mythos to AI Defense as Splunk Tracks AI Agent ROI

NewsTree24 | November 15, 2026 — Cisco announced on November 9, alongside Cisco Live APJC in Melbourne, that it plans to add Anthropic’s Claude Mythos to its AI Defense red-teaming framework. The company also outlined Splunk observability updates intended to help enterprises monitor AI-agent activity, assess output quality and connect usage costs with business value.

What Cisco is adding to AI Defense

Cisco says Claude Mythos will help its adaptive red-teaming framework design contextual, multi-step tests against an organisation’s registered AI applications and agents. The aim is to uncover weaknesses before those systems are put into production or exposed to real attackers.

The access model is deliberately constrained. According to Cisco, customers will interact through AI Defense’s structured interface rather than open-ended prompts. Testing is limited to targets the customer has registered, owns and is authorised to assess. Each engagement is tied to a named operator, logged for audit, time-limited and stoppable by the customer.

These safeguards are important because a model capable of generating sophisticated attack plans could be misused if it were exposed without controls. Cisco says the integration is intended for defensive testing, not unrestricted scanning of internet systems.

Splunk Tokenomics links agent costs with evaluation results

Cisco also described Tokenomics, an enhancement to Splunk Agent Observability. It is designed to combine usage data from Amazon Bedrock and Amazon CloudWatch with evaluations of an agent’s work, giving teams a view of spending alongside output quality and impact. The goal is to help finance, engineering and security teams see which agents consume resources and whether the work they produce is useful.

Cisco said Tokenomics would enter private preview in December 2026 and reach general availability through AWS Marketplace in early 2027. It is not generally available at the time of this report. The company said its Eval-G1 evaluation models and Evaluation Tuner would be generally available starting in November.

Why AI-agent security is becoming an enterprise issue

Traditional software security focuses on code, identities, endpoints and network activity. AI agents add another layer: they can interpret instructions, call tools, query data and take multi-step actions. A weakness in an agent’s prompt handling or tool permissions can therefore turn into a data-access or workflow problem.

Red teaming and observability address different parts of that risk. Red teaming tests how a system may fail before or during deployment. Observability helps teams understand what happened during actual use, including tool calls, latency, token consumption and output quality. Neither replaces identity controls, least-privilege permissions, human approval for high-impact actions or incident-response planning.

Questions security and IT leaders should ask

  • Can testing be restricted to approved systems and specific environments?
  • Are prompts, tool calls, operator identities and test results retained in an audit trail?
  • Can an agent’s access be revoked quickly if it behaves unexpectedly?
  • Does cost reporting include model usage, infrastructure and downstream tool costs?
  • Are evaluation metrics tied to business outcomes rather than token volume alone?

Availability: what is confirmed

Cisco’s November 9 announcement says the Claude Mythos red-teaming integration is coming soon for select enterprises. Tokenomics is planned for private preview in December and AWS Marketplace availability in early 2027. Cisco says Eval-G1 evaluation models and the Evaluation Tuner are available starting this month. Organisations should confirm eligibility and release status with Cisco before making procurement or deployment decisions.

For related NewsTree24 reporting, read about CrowdStrike’s AI Kill Switch and WhiteMatter’s AI-agent security platform. More coverage is available in our AI Tools section.

Sources and attribution

Leave a Comment