Palo Alto Networks introduces XCOR to investigate incidents

The platform is designed to investigate software alerts and recommend fixes, while engineers remain involved by default.

The platform is designed to investigate software alerts and recommend fixes, while engineers remain involved by default.
Palo Alto Networks introduced Cortex XCOR last week to investigate software incidents and recommend fixes, rather than leave engineers to interpret monitoring dashboards and investigate manually. Engineers are still notified when incidents occur, the company’s observability executive Martin Mao told The New Stack.
XCOR is an observability platform—a tool for understanding how applications and the infrastructure supporting them are behaving. It was built by the Chronosphere team inside Palo Alto Networks, which acquired Chronosphere in January.
According to Mao, an alert automatically triggers a specialized superintelligence (SI) agent that investigates the underlying problem and recommends actions to fix it or reduce its impact. XCOR Operator provides a conversational interface to the agents working behind the scenes. Mao says specialized agents also handle tasks such as adjusting alerts and dashboards and optimizing data volumes.
The default workflow is “human-in-the-loop,” meaning engineers remain involved. Mao says the system pages an engineer when an incident occurs and can complete its investigation while that engineer logs in. Engineering leaders can expand the system’s permissions to act autonomously over time; automatic investigation does not, by itself, mean unattended fixes.
XCOR Fabric supplies the agents with context about applications, infrastructure and the organization. Its knowledge graph is described as a real-time model of infrastructure, applications and business logic. Its operational memory adds information from earlier incident investigations.
Fabric also draws on the queries senior engineers run, the dashboards they use, documentation and operational files. Runbooks—written procedures for handling operational problems—provide further human knowledge. These inputs give agents context beyond the alert itself.
The New Stack cites Mao’s company blog for the following results from the incident-investigation agent.
| Measure | Reported result |
|---|---|
| Average time to investigate and recommend actions | Under three minutes |
| Root-cause analysis success in complex production environments | 75% |
| Other incidents where the analysis was deemed useful | A further 19% |
The evaluation sample and measurement method remain unclear; these are company-reported figures, not independent test results. The reported investigation time is not a measure of how long it takes to fix an incident.
Palo Alto Networks announced its intention in July to acquire Embrace, a company that monitors real users’ experience of applications. It plans to combine that front-end monitoring with its in-house XCOR Synthetics and monitoring of backend systems and infrastructure. The company describes this combination as a planned route to visibility across the application and its supporting systems, rather than a delivered integration.
Pricing and availability details for Cortex XCOR remain unclear.

The features extend shopping beyond the TikTok Shop marketplace, but availability and terms remain unclear.
2 min

The Register's account of OpenAI's technical evaluation reports lower detection in shorter passages and weaker results for math or code. A missing watermark does not prove human authorship.
4 min

Internal documents describe a pre-launch effort to restrict the agent’s access to services and network destinations. An anonymous Meta source questioned whether the repairs were adequate.
3 min