Cisco Enhances AI Security with New Red Teaming Tools and Mythos Integration

Cisco has unveiled new AI red teaming and agent monitoring tools designed to bolster the security of AI agents that enterprises are increasingly deploying. This proactive approach aims to identify vulnerabilities in AI systems before malicious actors can exploit them.

The new capabilities are part of Cisco AI Defense, the company's security platform for AI models and applications. Unlike traditional testing methods that rely on human testers, Cisco's algorithmic approach automates the probing of models and agents for weaknesses across more than 200 risk categories, including prompt injection and data exfiltration.

In a significant update, Cisco has integrated Anthropic's Mythos model into its AI Defense product. This partnership allows Cisco to enhance its adaptive red teaming capabilities, which simulate attacks on AI applications to uncover potential threats. Jeff Schultz, Cisco’s senior vice president for portfolio strategy and marketing, emphasized that AI Defense is designed to probe and break models, with Mythos aiding in the identification of more vulnerabilities.

Additionally, Cisco's platform employs a full lifecycle approach to AI security, following a discover-detect-defend sequence that includes real-time threat intelligence from Cisco Talos. This ensures that the testing remains relevant against current attack techniques.

As enterprises transition from basic chatbots to more sophisticated AI agents capable of taking actions, the need for robust security measures becomes paramount. Cisco's new tools aim to address these emerging risks, helping organizations maintain trust in their AI deployments.