Home TechnologyAmazon Uncovers AI-Driven Cyberattack That Exploited Medical Software in Minutes

Amazon Uncovers AI-Driven Cyberattack That Exploited Medical Software in Minutes

by Phoenix 24

Autonomous hacking agents expose a new cybersecurity threat.

SEATTLE

Amazon has revealed a cybersecurity operation in which autonomous artificial intelligence agents discovered a previously unknown vulnerability in medical imaging software and exploited it within approximately 11 minutes. The attack targeted university research infrastructure in East Asia and demonstrated how AI systems can independently identify security weaknesses with limited human direction. Amazon Threat Intelligence attributed the activity to a suspected China-linked actor with moderate confidence, while emphasizing that the evidence did not establish involvement by the Chinese government.

The investigation, disclosed on November 20, identified an operation conducted between September 8 and 10 using Cairn, an open-source framework designed to coordinate autonomous cybersecurity agents. The attackers supplied university internet addresses and instructed the system to obtain command-line access, leaving the agents to determine their own methods. Amazon discovered the campaign after finding the attackers’ Cairn management server publicly accessible without authentication, exposing operational records and instructions.

The breakthrough occurred against MONAI Label, an open-source application used by medical researchers to annotate diagnostic images and develop artificial intelligence models. The agents identified a vulnerability in how the software handled uploaded files, allowing them to write information outside the intended storage directory. Within approximately 11 minutes, they progressed from examining the application’s programming interface to demonstrating unauthorized file-writing capabilities.

The distinction between that initial achievement and full server compromise is important. Amazon reported that the attackers gained administrator-level control during a subsequent session, after exploiting the same vulnerability to execute unauthorized code. The successful intrusion affected a university server, while evidence suggesting access to another university network remained inconclusive.

The operation was not uniformly successful. In separate attempts, the autonomous agents spent approximately 16 hours testing 148 actions against a university database without achieving their objective. Another campaign involved more than 30,000 login attempts over six hours, also without success. These failures reveal that autonomous systems remain constrained by technical defenses, even as their ability to conduct continuous experimentation changes the speed and scale of cyberattacks.

Amazon’s attribution relied on instructions written in Simplified Chinese, computer timestamps configured to China Standard Time and the selection of universities considered strategically relevant to China. However, these indicators do not establish the attackers’ nationality or demonstrate state sponsorship. The company did not identify a specific hacking organization, nor did it report confirmed theft of patient records or medical images.

The discovered vulnerability affects MONAI Label version 0.8.5 and earlier releases and has been classified as critical. Although developers incorporated corrective changes into the source code in October, the project’s security advisory indicated that a patched software release was not yet available. Security recommendations include restricting access from untrusted networks, enabling authentication and limiting the privileges assigned to application processes.

The incident raises concerns beyond the immediate vulnerability. Medical research environments increasingly depend on interconnected platforms that process sensitive information and support the development of diagnostic technologies. A weakness in one specialized application can create opportunities for unauthorized access, particularly when research systems are exposed to public networks without adequate security controls.

More fundamentally, the case demonstrates a shift in the relationship between artificial intelligence and cybersecurity. Autonomous agents can accelerate defensive assessments, but similar capabilities can also enable malicious actors to investigate vulnerabilities continuously and adapt their strategies without constant human supervision. Organizations may consequently need faster detection, stronger preventive safeguards and automated defensive responses operating under appropriate human oversight.

Amazon’s findings do not establish a widespread compromise of healthcare systems. They do, however, illustrate how rapidly an overlooked software vulnerability can become an operational security incident when autonomous AI capabilities are used offensively. As artificial intelligence becomes more deeply integrated into research and healthcare infrastructure, maintaining trust will increasingly depend on whether security practices evolve as quickly as the technologies they are intended to protect.

Where Innovation Meets Responsibility.

You may also like