A threat actor utilized agentic artificial intelligence to compromise an Amazon Web Services (AWS) cloud environment within 72 hours, a task that would typically take weeks, according to a report by the security vendor Sygnia. The attack, aimed at extortion, leveraged familiar cloud infrastructure exploitation techniques but at an accelerated pace due to AI assistance.
The report, titled "Inside an AI-Assisted Cloud Attack: Familiar Techniques at Unfamiliar Speed," detailed how the attacker exploited control gaps in secrets management, identity governance, deployment workflows, and cloud permissions. The initial entry point was an internet-facing application that exposed an access key to one of the AWS accounts.
Once inside, the threat actor employed AI-assisted or agentic workflows to simultaneously pursue four key objectives. These included searching for and stealing secrets and credentials across various layers of the AWS environment, such as plaintext secrets in S3 buckets, API keys from application databases, and sensitive information stored in AWS Secrets Manager and AWS Systems Manager Parameter Store.
Concurrently, the attacker focused on establishing persistence mechanisms. This involved creating new access keys and Identity and Access Management (IAM) users, setting up reverse shells on EC2 instances and ECS containers, and altering deployment files to maintain access and control.
Data exfiltration was another critical objective, with the attacker targeting data stored in Relational Database Service (RDS) databases. The final phase involved impact actions designed to demonstrate their capabilities to the victim organization.
These impact actions included denying access to S3 buckets, reducing ECS services or containers to zero capacity, implementing Access Control List (ACL) rules to block network access, and purging Simple Queue Service (SQS) queues.
Sygnia's report highlighted that the organization's existing gaps in visibility, monitoring, identity controls, and incident preparedness significantly aided the attacker's rapid progress.
Avi Dayan, VP of incident response at Sygnia, emphasized that the most striking aspect of the attack was the speed and volume of malicious activity executed in a compressed timeframe post-intrusion. He noted that the increasing accessibility of large language models and agentic AI lowers the barrier to entry for threat actors, enabling them to operate with unprecedented speed and scale, even for those with fewer resources or less sophistication.
To mitigate such threats, Sygnia recommended several containment measures for network defenders. These include restricting cloud management access through IP allowlisting, disabling remote access VPN connectivity until containment is complete, and limiting outbound internet connectivity for cloud resources to approved destinations. Additionally, applying firewall policies and network ACLs to block malicious infrastructure and unknown assets, enforcing IP restrictions on code repositories, routing all application traffic through web application firewalls (WAFs), and implementing network segmentation and isolation controls were advised.
