APERION (previously LangSmart), the enterprise AI governance firm, introduced the launch of the SmartFlow SDK, offering a safe, on-premises path for enterprises migrating away from compromised cloud-based AI gateways. The launch coincides with a 200% improve in net site visitors for the reason that March 24 LiteLLM provide chain assault that compromised an estimated 36% of all cloud environments.
Additionally Learn: AiThority Interview with Glenn Jocher, Founder & CEO, Ultralytics
“Essentially the most extensively used open-source LLM proxy simply acquired supply-chain attacked. 36% of cloud environments affected. SmartFlow was unaffected. The deployment mannequin is the safety mannequin.” — Craig Alberino, CEO, APERION
LiteLLM was the sufferer of a provide chain assault during which the menace actor group TeamPCP compromised essentially the most extensively used open-source LLM proxy within the Python ecosystem by means of a cascading breach of Aqua Safety’s Trivy vulnerability scanner. The compromise has pressured enterprises throughout monetary companies, healthcare, and protection to urgently reassess their AI infrastructure dependencies. Your complete LiteLLM bundle, with roughly 95 million month-to-month downloads, stays quarantined on the Python Package deal Index.
APERION’s flagship product, SmartFlow, is an on-premises, Kubernetes-native AI governance management aircraft that was architecturally unaffected by the incident. SmartFlow is deployed as a software program equipment behind the enterprise firewall, with no dependency on public bundle registries, no cloud knowledge transit, and no exterior CI/CD pipelines within the buyer deployment path.
“The LiteLLM provide chain assault is the AI period’s SolarWinds or NotPetya second. It validates what we’ve been constructing towards since day one: regulated enterprises can not govern their AI from the cloud,” mentioned Craig Alberino, CEO and Co-Founding father of APERION. “Within the week for the reason that breach, we’ve seen a 200% improve in net site visitors from enterprises looking for LiteLLM alternate options and AI gateway safety. These are usually not startups. These are establishments that outline what production-grade means in monetary companies and healthcare. AI governance infrastructure is an estimated $40 to $50 billion market alternative, and enterprises at the moment are making buying selections primarily based on deployment mannequin and provide chain safety, not simply options.”
SmartFlow is in manufacturing with paying enterprise prospects together with DDA, the main AI-powered business actual property funding due diligence platform, which has achieved 99.999% uptime over 4 months of steady operation. Lively evaluations are underway at a number of Fortune 500 establishments in monetary companies. The corporate holds 22 patent positions overlaying enterprise AI governance, sovereign mannequin deployment, and autonomous AI management aircraft structure.
The SmartFlow SDK, additionally launched this week, is a Python library offering enterprise builders a direct path from analysis to production-grade AI governance. The SDK detects whether or not a SmartFlow equipment is accessible and configures accordingly: full enterprise options with an equipment, or a standalone software program gateway with function parity to LiteLLM and OpenRouter with out one. APERION has additionally revealed a migration whitepaper for organizations transitioning from compromised or discontinued AI gateway dependencies.
“The March 2026 provide chain assault was not an anomaly. LiteLLM had 17 or extra CVEs earlier than this incident,” mentioned Scott Ancheta, CTO and Co-Founding father of APERION. “When your AI governance layer is determined by a public bundle registry and an unaudited CI/CD pipeline, you aren’t operating enterprise-grade infrastructure. SmartFlow was constructed from the bottom up as a ruled equipment as a result of we understood that the deployment mannequin is the safety mannequin. Our Rust-based infrastructure delivers sub-5 millisecond routing overhead and our MetaCache semantic caching achieves 55% to 75% hit charges at p95 latency on manufacturing {hardware}. These are revealed, verifiable benchmarks from NVIDIA GTC 2026.”
Additionally Learn: The Infrastructure Warfare Behind the AI Increase
[To share your insights with us, please write to psen@itechseries.com]
