CrowdStrike adds an emergency stop for rogue AI agents and expands agent security through acquisitions

What changed: CrowdStrike introduced an AI Kill Switch in Falcon Guardian that lets authorized security staff terminate a running AI agent, cut its network access, and prevent it from relaunching. The company also agreed to acquire Fiddler AI, whose software monitors agent behavior and supports evaluation, traceability, and explainability; Axios reported the Fiddler deal at approximately $400 million.

Why it matters: Companies deploying agents with access to production systems should treat interruption as a core control, not an afterthought. A kill switch outside the agent’s decision process can limit damage from prompt injection, runaway automation, or unauthorized actions, while observability tools can help explain what happened afterward.

Try/watch: Add a human-triggered stop procedure to every agent rollout, test it under realistic failure conditions, and verify whether the control can also revoke credentials and isolate downstream tools.

Palo Alto Networks reportedly targets Clutch in a $500 million AI-agent security deal

What changed: Palo Alto Networks is reportedly in talks to acquire Clutch Security for about $500 million in cash, plus possible milestone payments. Clutch focuses on identifying and controlling AI agents’ access to enterprise systems; the reported discussions have not produced a signed agreement, and neither company publicly confirmed the terms.

Why it matters: The potential price signals that agent identity, permissions, and access monitoring are becoming strategic cybersecurity capabilities rather than niche developer tooling. Buyers should expect agent security to converge with identity management, cloud security, and privileged-access controls—and should avoid treating an agent as merely another API client.

Try/watch: Inventory every autonomous system operating in the organization, record which data and tools each can reach, and require narrowly scoped permissions with a clear owner and expiration date.

KODIF and Cavalry combine AI agents with post-purchase execution for ecommerce

What changed: KODIF and Cavalry announced a partnership to build an agentic post-purchase platform for ecommerce and said they had secured additional financing from Bling Capital, SNR, Mistral, and Red Bike Capital. The planned system combines KODIF’s customer-service agents—which interpret intent and reason through policies—with Cavalry’s software and execution infrastructure for resolving issues after checkout. The companies said total funding across the two businesses reached $19 million.

Why it matters: This is a practical shift from agents that only answer questions to agents that can complete operational work such as resolving customer conversations and executing policy-driven outcomes. Merchants evaluating similar systems should focus less on conversational quality alone and more on whether the agent can safely perform refunds, replacements, shipping changes, and escalations with an auditable trail.

Try/watch: Pilot one narrowly defined post-purchase workflow, set monetary and policy limits, and measure resolution accuracy, unauthorized concessions, escalation rates, and customer satisfaction before expanding the agent’s authority.