CyberStrikeAI Deploys 44 AI Agents to Target Japanese Organizations

CyberStrikeAI Deploys 44 AI Agents to Target Japanese Organizations

CyberStrikeAI, an open-source penetration testing framework, was employed in October 2026 to conduct automated attacks against 19 organizations, 16 of which were located in Japan. The campaign leveraged 44 AI agents that coordinated reconnaissance, vulnerability discovery, exploitation, command-and-control, persistence, data exfiltration, and automated reporting.

Attack Timeline and Workflow

1. Reconnaissance – Agents scanned target networks, collected system inventories, and identified vulnerable services.
2. Vulnerability Validation – Automated scripts verified CVEs and misconfigurations, logging findings in a structured format.
3. Exploitation – Exploit modules were triggered against validated weaknesses, establishing footholds.
4. Command-and-Control – A distributed C2 layer routed traffic through encrypted tunnels, minimizing detection.
5. Persistence – Agents deployed backdoors and scheduled tasks to maintain access.
6. Data Exfiltration – Collected data was compressed, encrypted, and sent to staging servers.
7. Reporting – Each target received a standardized report containing evidence, logs, and a summary of actions performed.

The entire workflow was orchestrated by the CyberStrikeAI platform, which delegated tasks among the 44 agents and generated Japanese-language fake login pages to facilitate credential harvesting.

Evidence and Attribution

TrendAI identified three distinct server environments used by the threat actors. Two environments were largely manual, operated by skilled human operators. The third environment matched the CyberStrikeAI architecture, with automated agent coordination.

Trend Micro’s enterprise security division confirmed that 19 validation activities were recorded, each producing a report in the same format. The reports included command histories, logs, and evidence indexes, indicating a repeatable, automated process rather than ad-hoc notes.

A second campaign, described by TrendAI, involved a human-led operation that employed a large language model to accelerate planning and tool development. Infrastructure analysis suggests that this campaign also targeted AI workflow platforms, though the exact scope remains unconfirmed.

Affected Products and Versions

The attacks did not target a single product family. Instead, the automated reconnaissance modules scanned for a broad range of vulnerabilities across operating systems, web applications, and network devices. No specific CVE identifiers were disclosed in the public analysis, but the methodology aligns with known exploitation patterns for common misconfigurations and unpatched software.

Detection and Containment Steps

1. Network Segmentation – Isolate critical assets and enforce strict egress controls.
2. Endpoint Detection – Deploy EDR solutions that flag anomalous agent behavior, such as repeated credential-harvesting scripts or unusual outbound traffic patterns.
3. Log Correlation – Enable centralized logging for authentication events, file integrity, and process creation. Correlate logs against the known command patterns from the CyberStrikeAI reports.
4. C2 Detection – Monitor for encrypted tunnels or unusual DNS queries that match the C2 signatures identified in the reports.
5. Patch Management – Apply the latest security updates for all software and firmware; prioritize fixes for CVEs commonly exploited by automated tools.
6. Threat Intelligence Feeds – Subscribe to feeds that include indicators of compromise (IOCs) from the CyberStrikeAI campaign, such as file hashes, IP ranges, and domain names.
7. Incident Response Playbook – Prepare a playbook that includes isolation, evidence preservation, and coordination with law enforcement, given the scale of the attack.

Monitoring for Hybrid Human-AI Operations

The second campaign’s use of a large language model to accelerate planning suggests that future attacks may blend human creativity with AI automation. Security teams should monitor for anomalous tool development activity, such as rapid creation of custom scripts or unusual usage of AI APIs.

Next Steps for Security Teams

• Review current network segmentation and ensure that critical assets are not exposed to broad reconnaissance.
• Verify that all endpoints run the latest security patches and that EDR solutions are configured to detect automated agent behavior.
• Integrate the CyberStrikeAI IOCs into SIEM rules to trigger alerts on matching patterns.
• Conduct tabletop exercises that simulate an automated attack chain to validate response readiness.

For further technical guidance, see the OpenAI AI Agents Exfiltrate via VM Escapes and Cross-Sandbox Channels article, which discusses similar automated exfiltration techniques.

Related coverage

• MythicRatatoskr Backdoor Uses Yggdrasil Mesh to Hide C&C, Enables Sandworm Data-Wipe
• Magento Stores Hit by Scanner-Evasion JavaScript Loader Campaign

Explore more on this topic

• More cyberattack coverage