2026年5月4日の
セキュリティ情報

Subject: [Info] Addressing Shadow AI with AI-BOMs

Hi team,

I am sharing information regarding the management of AI assets within our supply chain.

■ Overview
Traditional SBOMs are no longer sufficient for AI-infused environments. The emergence of AI-BOMs allows organizations to gain visibility into AI-specific components—including models, datasets, SDKs, and prompts—to mitigate the risks associated with 'Shadow AI.'

■ Scope
- All environments deploying AI applications, LLMs, or AI agents.

■ Recommended Steps
1. Conduct an inventory of all AI tools and libraries currently in use across the organization.
2. Evaluate methods to implement AI-BOMs to track model provenance and dependencies.
3. Establish monitoring processes to detect and govern unauthorized AI usage (Shadow AI).

■ Reference
- The Register: Shadow IT has given way to shadow AI. Enter AI-BOMs

Priority: Medium
Deadline: Next security review cycle

Subject: [Threat Intel] Technical Analysis of Python-based RAT 'Deep#Door'

Dear Security Team,

We are sharing technical details regarding a newly identified remote access trojan (RAT) named 'Deep#Door'.

■ Overview
Deep#Door is a Python-based malware that utilizes a single batch file (install_obf.bat) to deploy its payload, bypassing initial network-based detection. It employs sophisticated evasion techniques, including disabling Windows Defender, patching AMSI/ETW, and ntdll unhooking, to maintain long-term persistence.

■ Scope
- Windows Operating Systems

■ Mitigation Steps
1. Monitor for suspicious file creation in the %LOCALAPPDATA%\SystemServices\ directory via EDR.
2. Implement detection rules for unusual Python process behavior, specifically regarding registry modifications and scheduled task creation.
3. Enforce policies prohibiting the execution of untrusted batch files.

■ Reference
- Securonix Threat Research Team Analysis

Priority: High
Deadline: Immediate review

Subject: [Intel] AI-Driven Vulnerability Research via Anthropic Mythos

Dear Security Team,

We are sharing intelligence regarding the operational use of AI for vulnerability discovery.

■ Overview
Reports suggest that the US NSA is utilizing Anthropic's 'Mythos' AI model to identify security flaws in commercial software, specifically targeting Microsoft products. The ability of AI to conduct industrial-scale vulnerability hunting significantly increases the risk of rapid zero-day discovery.

■ Scope
- Microsoft products and other widely used commercial software.

■ Recommended Actions
1. Ensure all critical systems and applications are updated with the latest security patches.
2. Enhance monitoring via EDR/XDR to detect potential exploitation of unknown vulnerabilities.
3. Review and update the Incident Response plan to account for an increase in AI-driven threats.

■ Reference
- Bloomberg Report (April 30, 2026)

Priority: Medium
Deadline: Ongoing