Story perspectives

Homeland Security Orders Agencies to Stop Using DeepSeek AI

By Drooid · Oct 21, 2026, 10:28 AM UTC · How we work

Summary

Story summary

DeepSeek open-sourced its DeepInfer inference framework on October 20, 2026.
DeepInfer powers V4.1-Flash inference.
A DeepSeek cluster processes about 155 trillion tokens per day with DeepInfer.
DeepInfer runs on NVIDIA GPUs and Huawei Ascend NPUs, supporting distributed inference and speculative decoding.
The U.S. Department of Homeland Security ordered federal civilian agencies to stop using DeepSeek AI tools.

History

Historical Context

OpenAI announced GPT-2 in February 2019 and delayed the 1.5-billion-parameter model’s full release, citing misuse risks such as spam and fake news. The Allen Institute for AI responded with a detection tool, while the model’s partial release spurred open-source copies like OpenGPT-2, whose cloud-compute cost was about $50,000. These actions shaped broader safety debates.

1 of 2

Story summary

DeepSeek open-sourced its DeepInfer inference framework on October 20, 2026.
DeepInfer powers V4.1-Flash inference.
A DeepSeek cluster processes about 155 trillion tokens per day with DeepInfer.
DeepInfer runs on NVIDIA GPUs and Huawei Ascend NPUs, supporting distributed inference and speculative decoding.
The U.S. Department of Homeland Security ordered federal civilian agencies to stop using DeepSeek AI tools.

1 / 2