2026年4月22日の
セキュリティ情報

Subject: [Security Advisory] Risks of "Toxic Combinations" in AI Agent Permissions

Hi all,

This is a security update regarding a critical risk pattern emerging from AI agents and SaaS integrations.

■ Overview
Research highlights a risk known as "Toxic Combinations," where AI agents, MCP servers, or OAuth bridges link multiple applications. This creates a combined attack surface that is often invisible to single-application audits. A recent breach of the Moltbook database demonstrated this, resulting in the leak of 1.5 million API tokens and OpenAI keys due to unencrypted storage of cross-app credentials.

■ Scope
- Environments utilizing AI agents, bots, or MCP servers.
- Organizations with extensive SaaS-to-SaaS integrations via API or OAuth.

■ Recommended Actions
1. Establish a comprehensive inventory of all non-human entities (AI agents, bots, third-party connectors).
2. Shift audit focus from single-app permissions to "permission chains" across integrated services.
3. Apply the Principle of Least Privilege (PoLP) to API scopes and OAuth authorizations to eliminate over-privileged accounts.

■ Reference
- Cloud Security Alliance (CSA) 2025 SaaS Security State Report

Priority: High (Prompt review is recommended for organizations deploying AI agents or complex SaaS ecosystems).

Subject: [Security Advisory] Mirai Botnet Exploiting Discontinued D-Link Routers

Hi all,

This is a security notification regarding a critical vulnerability affecting legacy D-Link routers.

■ Overview
Akamai has reported that the Mirai botnet is actively exploiting a command injection vulnerability (CVE-2025-29635) in discontinued D-Link routers. Attackers can execute arbitrary shell scripts via crafted POST requests due to a lack of validation in a controllable function value.

■ Affected Scope
- Product: D-Link DIR-823X series
- Firmware Versions: 240126 and 24082

■ Required Actions
1. Promptly audit your network infrastructure to identify any active DIR-823X series routers.
2. Since these devices are end-of-life (EOL) and no security updates are available, please prioritize replacing them with secure, supported hardware and decommission the affected units.

■ Reference
- CVE-2025-29635
- Akamai Security Report

Priority: High (Prompt replacement is strongly recommended)

Subject: [Security Advisory] Risk of "Toxic Combinations" in AI Agent Cross-App Permissions

Hi all,

This is a security notification regarding the risks associated with permission stacking in AI agent integrations.

■ Overview
Researchers disclosed a data leak in "Moltbook," a social network for AI agents, where misconfigured databases exposed API tokens and plaintext third-party credentials (e.g., OpenAI API keys). This highlights a vulnerability known as "Toxic Combinations," where permissions across multiple applications stack via an AI agent or OAuth grant, creating a risk surface that is often invisible during standard single-app access reviews.

■ Scope
- Environments utilizing AI agents, OAuth integrations, and cross-app automation.
- SaaS applications storing credentials in plaintext.

■ Recommended Actions
1. Audit permissions granted to AI agents and integration tools to ensure the principle of least privilege is applied.
2. Ensure that API keys and secrets are encrypted or managed via dedicated secret management tools rather than stored in plaintext.
3. Implement a review process that visualizes the "chain of permissions" across integrated apps, rather than reviewing apps in isolation.

■ Reference
- The Hacker News: Toxic Combinations: When Cross-App Permissions Stack into Risk

Priority: High (Prompt review and mitigation are recommended)

Subject: [Action Required] Critical Privilege Escalation Vulnerability in ASP.NET Core (CVE-2026-40372)

Hi all,

This is a security advisory regarding a critical vulnerability identified in ASP.NET Core.

■ Overview
An improper verification of cryptographic signatures in ASP.NET Core allows an unauthorized attacker to elevate privileges over a network. With a CVSS score of 9.1, a successful exploit could grant the attacker SYSTEM privileges, leading to unauthorized file disclosure and data modification.

■ Affected Scope
- Applications using the Microsoft.AspNetCore.DataProtection 10.0.6 NuGet package (either directly or via dependencies such as Microsoft.AspNetCore.DataProtection.StackExchangeRedis).

■ Remediation Steps
1. Audit your project dependencies to identify the use of the affected NuGet package version.
2. Update the library to the latest patched version and redeploy the affected applications.

■ Reference
- Microsoft Security Advisory
- CVE-2026-40372

Priority: High (Prompt remediation is strongly recommended)

Subject: [Security Notice] Beware of Fake Cryptocurrency Wallet Apps

Hi everyone,

A series of phishing apps known as "FakeWallet" have been discovered in the Chinese App Store. These apps impersonate popular cryptocurrency wallets (such as MetaMask and Coinbase) to steal seed phrases and assets.

How to protect yourself:
1. Never install apps from unofficial sources or accept suspicious "provisioning profiles" on your iOS device.
2. Carefully check the app name for typos and verify the developer's identity before downloading.
3. Never enter your seed phrase or recovery phrase into any app or website unless you are absolutely certain of its authenticity.

Please remain vigilant to ensure the security of your digital assets.

件名: 【共有】iOS向け仮想通貨ウォレットを装ったフィッシングキャンペーン（FakeWallet）について

お疲れさまです。標記の脅威に関する情報共有です。

■ 概要
中国のApp Storeにて、MetaMaskやCoinbase等の人気ウォレットを装った26個のフィッシングアプリ（FakeWallet）が検出されました。攻撃者はタイポスクワッティングや偽の計算機アプリ等を用いてユーザーを誘導し、iOSのエンタープライズプロビジョニングプロファイルを悪用してトロイの木馬をサイドロードさせ、シードフレーズを窃取します。

■ 影響範囲
- 中国国内のiOSユーザーおよび仮想通貨ウォレット利用者

■ 対応手順
1. 社員に対し、公式ストア以外からのプロビジョニングプロファイル導入の危険性を周知してください。
2. MDM（モバイルデバイス管理）を導入している場合は、未承認のプロファイルインストールが制限されているか確認してください。
3. 不審なアプリのインストールが確認された端末の隔離および調査を検討してください。

■ 参考情報
- Kaspersky / xakep (FakeWallet campaign)

対応優先度: 中（ユーザーへの注意喚起を優先的にご検討ください）

Subject: [FYI] Phishing Campaign Targeting Crypto Wallets on iOS (FakeWallet)

Hi,

This is a security advisory regarding a phishing campaign targeting iOS users in China.

■ Overview
Twenty-six phishing applications, collectively named "FakeWallet," were discovered in the Chinese App Store. These apps impersonate well-known wallets (MetaMask, Coinbase, etc.) using typosquatting and deceptive app categories (e.g., calculators). The attackers leverage iOS enterprise provisioning profiles to sideload trojanized versions of wallets to steal users' seed phrases.

■ Scope
- iOS users in China and cryptocurrency wallet users.

■ Recommended Actions
1. Alert employees about the risks of installing provisioning profiles from untrusted sources.
2. If using MDM (Mobile Device Management), verify that the installation of unauthorized configuration profiles is restricted.
3. Implement monitoring or auditing for unauthorized app sideloading on corporate-managed mobile devices.

■ Reference
- Kaspersky / xakep (FakeWallet campaign)

Priority: Medium (Prompt user awareness is recommended)