The short answer
- On November 13, 2026, Amazon said its MadPot decoy sensors captured requests from AI agents between October 3 and November 5, 2026, and that the requests carried the agents' own records of campaigns run on ARTEX.
- Amazon Threat Intelligence wrote that it "cannot verify the outcomes the agents reported." It assesses with moderate confidence that ARTEX was used, and with moderate confidence that the operators were disparate Chinese-nexus threat groups.
- According to the agents' records, as Amazon describes them, the main campaign was a self-sustaining theft of credentials to access AI models, in which stolen keys paid for the inference that powered further key hunting.
- Amazon says the agents focused on internet-exposed LLM gateways and proxy platforms, including One API, New API, Veloera, Sub2API, and LiteLLM, and AI application builders including Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat, and Ollama.
- In a second campaign described in the same records, Amazon says agents attempted a know-your-customer (KYC) bypass with deepfakes at two cryptocurrency exchanges, and that the vendor's server-side verdict rejected the deepfake in every attempt the agents recorded.
- Amazon notified potentially affected AWS customers and published indicators of compromise in its post. It says getting security fundamentals right remains the most effective defense.
Amazon said on November 13, 2026 that its MadPot decoy sensors captured requests from AI agents between October 3 and November 5, 2026, and that those requests carried the agents' own records of campaigns run on ARTEX, an agentic AI penetration testing framework. In an AWS Security Blog post dated November 13, 2026, CJ Moses, CISO of Amazon Integrated Security, described what Amazon Threat Intelligence learned from those records.
Amazon wrote: "Amazon Threat Intelligence did not observe that activity as it happened and cannot verify the outcomes the agents reported, but the records show what the actors were pursuing." The post says Amazon Threat Intelligence "assessed threat activity referencing potentially affected AWS customers and notified them as appropriate." Amazon also published indicators of compromise in the same post, so organizations that run the affected technologies can review their own environments.
#What did Amazon observe?
Amazon says that from October 3 to November 5, 2026, MadPot, "a globally distributed network of decoy threat sensors", captured requests from AI agents that included operating memory from offensive campaigns that threat actors ran on ARTEX. The company says the agents sent those requests to MadPot sensors "configured to present as popular large language model (LLM) inference APIs", and that the requests carried "the agents’ own records of their campaigns."
Amazon Threat Intelligence "assesses with moderate confidence that ARTEX was used for these campaigns due to prompt references to ARTEX tooling and other technical artifacts referencing ARTEX." It also assesses, with moderate confidence, that the operators were disparate Chinese-nexus threat groups. Amazon cites the use of simplified Chinese language in prompts, China time-zone configurations, targeting it describes as consistent with Chinese interests, and tooling it says is predominantly used by Chinese-nexus threats.
The Hacker News, in a November 16, 2026 report by Swati Khandelwal, also covered the Amazon post. The Hacker News reported that the FBI said in a FLASH alert that a range of attackers have used ARTEX since at least August against organizations in many sectors around the world.
#What do the agents' records describe?
According to the agents' records, as Amazon describes them, "The most prominent ARTEX campaign was a large-scale theft of credentials to access AI models, organized as a self-sustaining operation in which stolen keys paid for the LLM inference powering further credential hunting." Amazon wrote: "The agents’ records show stolen keys being fed into the framework’s own model gateway, where they powered the agents searching for more keys."
Amazon says the agents used a shared method: "find internet-exposed AI applications, extract the upstream provider keys they hold, validate those keys, and store them in a managed pool." Amazon wrote: "The agents focused on applications that aggregate provider keys, because one weak management interface can expose many downstream accounts." The post states: "Targets included LLM gateways and proxy platforms such as One API, New API, Veloera, Sub2API, and LiteLLM, and AI application builders such as Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat, and Ollama."
Amazon describes a playbook that mapped each fingerprinted product to a known attack chain, including uninitialized installs that could be claimed, anonymous management and provider-configuration endpoints, default administrator credentials such as the LiteLLM default master key, CVE-2025-3248 in Langflow, CVE-2026-28797 in RAGFlow, an authentication bypass in One API family systems, and secrets left in front-end code and configuration files.
To find leaked keys, Amazon says, discovery drew on internet scanning services, certificate transparency, and public code and artifact sources, including GitHub repositories and event archives, npm and PyPI tarballs, container image layers, notebooks, model-hosting spaces, paste sites, and Telegram channels. "Keys exposed publicly for more than 30 days were treated as low-value inventory." "Beyond AI model access keys, cloud credentials were treated as a doorway to more access." Amazon says the agents prioritized access to premium frontier models over general cloud permissions. The records, Amazon says, also described reconnaissance, vulnerability testing, credential theft, cloud-privilege expansion, and cryptocurrency fraud attempts "across hundreds of targets spanning AI application platforms and other workloads."
A second campaign, Amazon says, "targeted cryptocurrency exchanges for know-your-customer (KYC) verification bypass using real-time deepfake injection." Amazon says threat actors used AI agents in an attempt to get past the identity verification checks at two cryptocurrency exchanges with deepfakes, and wrote that "the vendor’s server-side verdict rejected the deepfake in every attempt the agents recorded." Amazon says at least four agent sessions ran at once: one searched for ways to cash out, one worked on clearing account balances, one set up the deepfake video pipeline, and one carried out the verification bypass.
#What should operators do?
Amazon says organizations should assume exposed secrets and misconfigurations will be found quickly and remediate accordingly, and wrote: "Getting those fundamentals right remains the most effective defense." It also wrote: "None of the techniques in these campaigns were new. What AI agents change is the speed and persistence of the attacker." Amazon wrote: "The agents succeeded where basic security hygiene was missing, including default or hardcoded credentials, keys left in code and public artifacts, unauthenticated interfaces and inference endpoints, and unpatched software."
The company says organizations should bring AI applications, model gateways, and inference endpoints under the same controls as other production systems, and should watch for misuse of AI services through unexpected spend or unusual usage. Indicators of compromise are in Amazon's November 13, 2026 post.
The Hacker News wrote: "Change default administrator credentials, such as LiteLLM's default master key, and do not leave a new install unfinished where an outsider could take it over." It also wrote: "Search configuration files, environment variables, deployment scripts, CI/CD pipelines and databases for embedded secrets, keep scanning repositories and container images for exposed ones, and rotate exposed credentials quickly, as the FBI advises."
On the two named flaws, The Hacker News reported that Langflow versions before 1.3.0 are affected by CVE-2025-3248, and that 1.3.0 is the fixed version. For RAGFlow CVE-2026-28797, The Hacker News reported that the National Vulnerability Database (NVD) lists 0.24.0 and earlier as affected, that the advisory lists no patched version, and that version 0.25.0 "renders templates in the sandbox the advisory recommends".
#How does this compare?
Amazon's November 13, 2026 report is a separate account from earlier ARTEX coverage on DotsBot, including Microsoft's report that an AI agent reached Key Vault data in a five-minute case. A separate Microsoft report covers Storm-0940's use of ARTEX in a ConnectWise supply-chain attack. Operators of LiteLLM may also want our earlier report on LiteLLM's cosign signing-key rotation.
Our analysis: The picture Amazon presents is of AI agents that use stolen model keys to pay for more key hunting, plus a know-your-customer campaign in which, Amazon says, the vendor rejected the deepfake in every attempt the agents recorded. That picture comes from the agents' own records, and Amazon says it cannot verify the outcomes the agents reported. For operators of internet-exposed AI gateways and app builders, Amazon's practical point is that default credentials, uninitialized installs, leaked keys, unauthenticated management interfaces, and unpatched software are what the agents' playbook targeted.
Frequently asked questions
Did Amazon confirm that LLM API keys were stolen?
No. Amazon says the captured requests carried the agents' own records of their campaigns, and that it "cannot verify the outcomes the agents reported." The account of key theft in this article is Amazon's description of those records.
Which products does Amazon say the agents targeted?
Amazon lists LLM gateways and proxy platforms such as One API, New API, Veloera, Sub2API, and LiteLLM, and AI application builders such as Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat, and Ollama. Amazon describes these as products the agents targeted, not as products it says are flawed in every install.
What time period does the Amazon report cover?
Amazon says MadPot captured the requests from October 3 to November 5, 2026. The AWS Security Blog post, by CJ Moses, is dated November 13, 2026.
What should operators of these products do?
Amazon says organizations should assume exposed secrets and misconfigurations will be found quickly, bring AI applications, model gateways, and inference endpoints under the same controls as other production systems, and watch for unexpected spend or unusual usage. The Hacker News also advises changing default administrator credentials, such as LiteLLM's default master key, searching for embedded secrets, and rotating exposed credentials quickly.
What happened in the cryptocurrency exchange campaign?
Amazon says a second campaign used AI agents in an attempt to bypass know-your-customer (KYC) checks at two cryptocurrency exchanges with deepfakes. According to the agents' records, as Amazon describes them, the vendor's server-side verdict rejected the deepfake in every attempt the agents recorded.
Sources
Every factual claim in this article traces back to one of these.
One email a week on AI agents and bots.
Launches, pricing and rollout changes and the guides worth reading. At most once a week, and you can leave with one click.

