Loading...

Relying on low-cost AI inference providers and API routers to cut model costs can expose your operational infrastructure and credentials to third-party attackers. According to The Decoder, security firm Dragos sent roughly 2 million test requests containing fake credentials to AI providers and uncovered active data theft across multiple compromised intermediaries.
The Decoder reports that Dragos conducted testing from July through November 2026, identifying at least three compromised AI providers where attackers were actively harvesting user data. None of the three providers were aware of the breaches until Dragos notified them. In the largest incident, attackers harvested hundreds of AWS keys—including credentials from requests that the provider had rejected prior to processing—and tested them within hours to search S3 storage and establish persistent network access. Attackers deployed tools ranging from ARTEX, an AI-powered attack framework, to simple SSH login attempts from residential connections in France using passwords pulled directly from intercepted prompts.
AI routers like OpenRouter aggregate API calls and route prompts across multiple provider hops, such as Novita, AtlasCloud, StreamLake, and Phala. Dragos found that each intermediary hop in these open marketplaces can view the full request and response. Resellers frequently logged prompt data simply by setting the lowest price to attract traffic. Furthermore, privacy guarantees like Zero Data Retention (ZDR) were routinely ignored; a Chinese AI lab accessed secret test URLs embedded in more than half of all requests, even when processed through US providers. In one instance, a secret link appeared in an internal chat on Feishu (Lark) 24 days after a request and was accessed by six devices across Beijing and Shanghai. Following the report, OpenRouter removed three providers, revoked ZDR status for 125 endpoints, and notified users whose prompts were routed to MiniMax models. Providers including Novita, Phala, and Together have since published preliminary security reports.
For small business operations, routing API traffic through low-cost aggregators to save fractions of a cent per token introduces severe cybersecurity risks. Operational teams frequently feed database connections, system prompts, customer records, and API credentials into automated LLM workflows. If those requests pass through unvetted marketplace resellers, every intermediary can inspect, store, or leak your payload. As Dragos Chief Research Officer HD Moore noted, "where your prompts go is a decision, and if you don’t make it, the lowest price will make it for you." Using unmanaged cheap routing means your operational data and cloud storage keys could end up harvested by automated scripts or foreign networks before your IT team even realizes a breach has occurred.
Target is leveraging AI and advanced data science to make smarter inventory decisions, improving in-stock reliability for 90% of its assortment and enhancing delivery capabilities.

The Home Depot has expanded its flatbed delivery network, acquired distribution assets, and upgraded its order management software to better serve commercial contractors.

Starbucks reports record inventory availability with daily deliveries to almost 90% of its North American stores, after scrapping a nine‑month AI pilot and testing a 24‑hour model.

Amazon adds a 3.5% fuel surcharge to bulk storage services and lifts average FBA fees by $0.19 per unit, reshaping logistics costs for small sellers.

Comments