Google Threat Intelligence Group reported that threat actors used a multi-agent AI framework to compromise thousands of credentials in under six hours. Mandiant investigators linked the campaign to a suspected financially motivated actor. After breaking into an organisation’s cloud infrastructure, the attacker built an autonomous system from an AI coding chatbot, a prompt and agent instructions. Preconfigured markdown playbooks drove the scanning and harvesting. Troubleshooting and IP rotation ran without an operator. Traffic left the victim’s own addresses, making it look legitimate. The findings appear in the report “From Prompting to Autonomy: The Evolution of Adversarial AI,” which covers second-quarter activity. GTIG noted that adversaries are handing multistep decisions to models, shrinking the time defenders have to respond. The same report highlights related supply-chain activity by a group tracked as UNC6780, also known as TeamPCP.
Google Threat Intelligence Group reported that threat actors used a multi-agent AI framework to compromise thousands of credentials in under six hours. Mandiant investigators linked the campaign to a suspected financially motivated actor. After breaking into an organisation’s cloud infrastructure, the attacker built an autonomous system from an AI coding chatbot, a prompt and agent instructions. Preconfigured markdown playbooks drove the scanning and harvesting. Troubleshooting and IP rotation ran without an operator. Traffic left the victim’s own addresses, making it look legitimate. The findings appear in the report “From Prompting to Autonomy: The Evolution of Adversarial AI,” which covers second-quarter activity. GTIG noted that adversaries are handing multistep decisions to models, shrinking the time defenders have to respond. The same report highlights related supply-chain activity by a group tracked as UNC6780, also known as TeamPCP.