{
  "id": 16587724,
  "title": "Open-Source Models Are Ready for Enterprise…Here's How to Deploy Them Safely",
  "url": "https://urgent.news/2026/10/26/open-source-models-are-ready-for-enterprise-heres-how-to-deploy-them",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-26T03:14:01.000Z",
  "source": {
    "name": "HackerNoon",
    "slug": "hackernoon",
    "url": "https://hackernoon.com/open-source-models-are-ready-for-enterpriseheres-how-to-deploy-them-safely?source=rss"
  },
  "original_language": "en",
  "account": "A year ago, talking about DeepSeek or Qwen in an enterprise meeting would have met with silence. Now, those two models account for nearly 15% of the global AI market, up from just 1% at the beginning of 2025. The question has shifted from \"Are these models good enough?\" to \"How can we deploy them without creating a liability?\" Enterprises are more focused on the cost, with teams reporting an 80% reduction in inference bills compared to using only premium models. However, this cost advantage is contingent on proper deployment.\n\nThe skepticism surrounding open models is not unfounded. Early in 2025, Florida's state CFO banned DeepSeek, citing it as a national security risk, and several federal agencies followed suit. Security researchers discovered issues like weak encryption, potential SQL injection flaws, and undisclosed data transmissions in DeepSeek's R1 model. The issue is twofold: data potentially reaching the Chinese government and geopolitical hostility towards anything China-linked. The distinction is crucial, as bans often target the consumer-facing DeepSeek app running on Chinese servers, not the open-weight model used in-house.\n\nOrganizations that have successfully deployed open models have self-hosted MIT- or Apache-licensed models, ensuring no prompt data leaves their network. This approach eliminates most risks such as data transmission to foreign servers, opaque logging, and unclear retention. The remaining risk is a conventional vendor-risk conversation, focusing on model provenance, patch cadence, and access controls, similar to the due diligence applied to other software.\n\nHowever, self-hosting has its own challenges. It doesn't eliminate the risk of a single point of failure, but it does improve security by atomizing compute, model execution, and data handling. This strategy, as demonstrated by Aphanarc, prevents any single component from holding the complete picture of sensitive information. The calculus for CTOs has evolved. While some companies, like Coinbase and Cursor, have publicly adopted Chinese-origin models to cut costs, U.S. lawmakers are now scrutinizing the use of such models in American companies. The scrutiny is real, and the decisions enterprises make today will be the ones they must defend in the future.",
  "summary": "The key to safe deployment lies in self-hosting open models, controlling data transmission, and ensuring model provenance, patch cadence, and access controls.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}