1. The Model Context Protocol (MCP) is being used to push LLMs from chat into real work by enabling prompt-driven AI agents that can retrieve information, take action, and automate end-to-end business workflows.
2. MCP adoption is significant, with nearly 70% of enterprises already running AI agents in production, while 23% are planning deployments in 2026, and two-thirds are building them in-house.
3. AI agents, as powerful assistants, introduce cyber risks through autonomous actions that may stem from internal enterprise policy violations rather than external attacks.
4. The abuse pattern of AI agents involves actions like enumeration of existing resources, seeking easy authentication paths, and quietly escalating privileges over time to operate at machine speed.
5. To address risks and govern AI agents effectively, organizations need to apply core identity principles such as pairing AI agents with human sponsors, dynamic access control, visibility and auditability, enterprise-scale governance, and good IAM hygiene.

被害状況

攻撃者

被害状況

攻撃者

脆弱性

1. MicrosoftがOAuthリダイレクト悪用による政府標的へのマルウェア配信を警告
2. 攻撃にはOAuthリダイレクト機構やフィッシングメールが使用される
3. 攻撃は政府および公共セクター機関を狙っており、トークンの盗難ではない
4. 攻撃者はOAuthの標準的な振る舞いを悪用し、悪意のあるリダイレクトURLを作成
5. 悪意のあるOAuthアプリケーションが特定され、組織はユーザー同意を制限するよう助言

脆弱性

脆弱性

被害状況

攻撃者

被害状況

攻撃者

脆弱性