As enterprises increasingly adopt AI gateways to centralise access to foundation models, a new attack surface is emerging. A recent incident uncovered by Darktrace demonstrates that a compromised gateway can become a launchpad for far‑more damaging breaches.

Incident Overview

Researchers traced a threat actor’s entry into an EC2 instance that hosted an AI gateway linked to Amazon Bedrock. The attacker installed the XMRig cryptominer, turning the server into a mining node. Though the payload was relatively benign, the gateway’s broad permissions meant the attacker could have accessed sensitive prompts, API keys, or even pivoted deeper into the organization’s cloud stack.

Why Gateways Are a Prime Target

AI gateways sit at the intersection of identity, cloud infrastructure, proprietary data and multiple foundation models. By aggregating these capabilities, they become an attractive “mini supply chain” for adversaries. Once breached, an attacker can steal credentials, manipulate AI workflows or achieve persistence across AWS resources.

Potential Risks Beyond Mining

While cryptomining is noisy and easier to detect, more sophisticated objectives—such as credential theft, data exfiltration, or large‑scale AI‑inference abuse—could have had a far graver impact. Other vulnerabilities include model poisoning, prompt injection and insecure coding agents.

Mitigation Strategies

Experts recommend limiting IAM permissions, isolating gateway management interfaces from the internet, adopting short‑lived API keys, and segmenting AI infrastructure from production workloads. Continuous monitoring of AI‑specific administrative actions and prompt activity is also essential.

Looking Ahead

As AI becomes a core component of business workflows, treating gateways as privileged cloud assets and implementing rigorous controls will be vital. The most significant risks may stem from legitimate user interactions that inadvertently expose sensitive information.