Zentera Threat Briefing – May 2026

In this month's Threat Briefing, Nathanael Iversen covers three stories shaping the security outlook heading into May - from a major breach of HPE development environments, to a healthcare hack that hit over one million patients, to state-sponsored actors using AI agents to execute cyberattacks at machine speed.

Intel Broker Allegedly Breaches Hewlett Packard Enterprise

Threat actor Intel Broker claims to have breached HPE, with compromised assets reportedly including private GitHub repositories and Docker builds. Exposed repositories risk unauthorized access to proprietary code and intellectual property; compromised Docker builds pose risks to application deployment and security.

This incident is a reminder of the importance of segmenting critical assets and placing them behind Zero Trust access controls.

CHC Healthcare Hack Impacts Over One Million Patients

In a regulatory filing with the Maine Attorney General's office, CHC reported that a data breach impacted over one million individuals. Unusual activity was detected within CHC's systems, triggering an immediate cybersecurity investigation. A skilled attacker had accessed and extracted data within hours - without deleting or locking anything - before access was terminated. Daily operations were not disrupted, and CHC believes there is no ongoing threat.

The lesson: it doesn't take long for an experienced adversary to create expensive damage. Assume breach is not a cliché - it's a design requirement. Segmentation and least-privilege access must be in place before an attack occurs, not after. If your critical assets aren't in a chamber, now is the time.

China Uses Claude Code to Orchestrate AI-Powered Cyberattacks

State-sponsored threat actors from China used Anthropic's Claude Code and MCP tools to orchestrate automated cyberattacks at an unprecedented scale — using AI not just as an advisor, but to actually execute attacks. Claude Code acted as a central nervous system, processing operator instructions and breaking multi-stage attacks into discrete technical tasks offloaded to sub-agents. Roughly 30 targets were attempted across large tech companies, financial institutions, manufacturing companies, and government agencies. A subset of intrusions succeeded. Anthropic has since banned the relevant accounts and implemented defensive mechanisms to flag such activity.

Securing against agentic AI attacks requires controls at three layers: the endpoint, the infrastructure layer prior to LLM and MCP interactions, and the backend - limiting what agents and LLMs can reach.

Key Takeaways

• Segment critical assets and enforce Zero Trust access controls before a breach occurs - not after
• Assume breach is a design requirement; an experienced attacker can extract valuable data within hours
• Agentic AI attacks require enforcement at three layers: endpoint, pre-LLM infrastructure, and backend access controls
• If your critical assets aren't in a chamber, now is the time