Google: AI Agents Rescale Cyberattacks and Require Access Governance
Google Threat Intelligence Group technical lead Shane Huntley told iThome 台湾 that AI does not fundamentally change cyberattacks, but allows attackers to search vulnerabilities, analyze targets and execute existing techniques faster and at greater scale. He cited a possible multi-agent operation that could move from initial access to large-scale credential collection in under six hours, though the report does not provide supporting technical details. Google is also using AI agents for threat analysis while urging enterprises to apply least privilege, logging, auditing and human oversight.
AI agents can compress repetitive attack and defensive work, but enterprises also need controlled permissions and accountable human supervision so that autonomous tools do not become additional entry points.
圖片來源:
黃彥棻攝
一場網路攻擊,從取得受害系統的存取權限,到透過多個AI Agent協同執行攻擊,最後大規模蒐集存取憑證,可能不到六小時就能完成。攻擊者不再只是逐步尋找漏洞、手動執行攻擊,而是利用AI串聯不同階段的工作,在短時間內擴大攻擊規模。
Google威脅情報小組(Google Threat Intelligence Group,GTIG)技術長Shane Huntley指出,AI並未從根本上改變網路攻擊的本質,卻讓攻擊者能以更快的速度、更大的規模執行既有攻擊。過去受限於人力與時間,攻擊者必須選擇目標;如今AI可以協助搜尋漏洞、分析系統及執行攻擊,企業原本認為風險不高的弱點,也可能成為入侵入口。
當攻擊速度加快,防守方顯然不能再依賴人工逐一處理告警、分析漏洞及修補程式碼。Google正將AI導入威脅分析與防禦流程,運用大量AI Agent協助資安工作。不過,當Agent開始自主執行任務、存取企業內部系統,新的問題隨之出現:企業如何確保Agent只擁有完成任務所需的權限,而不會成為另一個遭到利用的攻擊入口?
#Google#Google Threat Intelligence Group#CodeMender#Google DeepMind#Okta#Comcast#IPIDEA