"Neither operation succeeded, yet together they show that the agents kept adapting for nearly a full day without human direction," Moses wrote.
Amazon CISO CJ Moses: discovery, dates, and immediate response
Amazon's CISO CJ Moses reported that Amazon’s Threat Intelligence Team discovered a publicly hosted instance of the open-source AI penetration-testing framework Cairn on September 10, 2026, providing visibility into a campaign that began on September 8 and ran for roughly 48 hours. The campaign targeted university networks in East Asia, outside Mainland China, Moses said, and the attacking prompts were written in Simplified Chinese with executed-script timestamps configured to China Standard Time. After identifying a zero-day in an application called MONAI Label, Amazon notified the project’s maintainers; they released a patch and a security advisory addressing the vulnerability.
How Cairn's AI agents operated autonomously
Moses described Cairn as an open-source, penetration-testing framework that has been adopted by criminal and espionage actors to run multi-stage intrusions with little or no human intervention. In this incident the agents carried out sequential actions, adapted when initial attempts failed, and shifted targets without continuous human commands. Amazon observed the agents mapping application frameworks, probing API endpoints, testing potential file-upload parameters for path traversal, and, in at least one session, using an exploit chain to write files and run a reverse shell as root.

Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleThe MONAI Label zero-day: path traversal to root shell in minutes
According to Moses, the agents identified MONAI Label’s framework, enumerated its API endpoints, and located an upload endpoint with a filename parameter that was vulnerable to path traversal. "The agents identified the application's framework, mapped its API endpoints, and found an upload endpoint with a filename parameter," Moses wrote. In about 11 minutes the agents went from mapping the application’s API to arbitrarily writing files on the server, and later they abused the same flaw to execute a Python reverse shell running as root. In a subsequent session a human operator supplied a proxy tunnel into a second university network and connected the agents to a second server running the same vulnerable code.
Earlier stages and failed techniques: Redis, Log4Shell, Qdrant, credential sprays
Before the MONAI Label exploit, the agents attempted a sustained attack against a university Redis server that ran roughly 16 hours and executed 148 sequential actions. Those actions included abusing Redis’s replication feature to load unauthorized code, escaping its scripting sandbox, writing to configuration files, and corrupting memory; those attempts did not succeed. The agents also tried a Log4Shell injection against a different service, an SSRF attack against a co-located Qdrant instance, and multiple credential-spray campaigns against PostgreSQL, SSH, and other services. Separately, another operation ran for more than six hours and targeted a university GPU server with credential‑spray attempts, compiling a list of 80 real user names gathered from public websites, academic papers and other documents. None of those earlier operations succeeded, Moses said, but they demonstrate the agents’ ability to iterate for extended periods without direction.
Gambit Security corroboration and the broader criminal usage of AI harnesses
The Amazon findings follow a Gambit Security threat-intelligence report that documented a China-linked operator using Cairn in September to carry out at least 105 attacks and compromise, "to varying degrees," at least 27 companies. Gambit reported that, in addition to Cairn, the actor used two other open-source AI harnesses to steal credit card records and other sensitive data from a Fortune 500 hospitality company, a major US airline, a large private US industrial supplies distributor, and a US online fashion retailer, among others. "Together, these cases show that this autonomous AI agent framework is being adopted in both espionage and criminal operations," Moses wrote.
What this means for university IT teams, open-source maintainers, and enterprise defenders
- University IT teams: incidents show the need to monitor for exposed, publicly hosted testing tools like Cairn and to prioritize rapid response to unusual, automated sequences of probing and exploitation attempts.
- Open-source maintainers (MONAI Label): maintainers were notified and released a fix and advisory; projects with upload endpoints and filename parameters should verify path traversal protections and implement prompt disclosure and patching procedures.
- Enterprise defenders: the Gambit findings and Amazon’s report together underline that AI-driven frameworks can be repurposed for both espionage and criminal theft, and that defenders must consider long probe periods, credential-spray detection, and protections for co-located services such as Qdrant and Redis.
The record Amazon assembled traces a short, adaptive campaign that ultimately exploited an application-level path traversal to gain root access, and it ties that activity to a broader pattern of AI-agent-enabled intrusions documented by Gambit Security. Amazon’s notification to maintainers and the subsequent patch for MONAI Label are concrete responses already taken; the incident leaves an open, practical question for defenders and maintainers alike: how many more exposed, automated agent instances are being used to test and then weaponize zero-days in widely adopted open-source projects?




