Close Menu
Stratnews GlobalStratnews Global
    Facebook X (Twitter) Instagram
    Trending
    • RWE pulls out of Namibia’s $10bn Hyphen green ammonia venture
    • Euclid Consortium Unveils Largest-Ever Universe Simulation with 3.4 Billion Galaxies
    • UN Arms Embargo Returns as Iran Warns of Retaliation
    • China’s Chang’e-6 Mission Wins IAF World Space Award at Sydney Congress
    • China’s Giant Antenna Powers Global Race for Lithium and Rare Earths
    • BRICS and IBSA Advance Space and Technology Cooperation amid Global Realignments
    • Judge Gives Preliminary Approval to $1.5B Anthropic Copyright Deal
    • Microsoft Blocks Israeli Military Use of Azure After Surveillance Findings
    Stratnews GlobalStratnews Global
    Write for Us
    Monday, September 29
    • Space
    • Science
    • AI and Robotics
    • Industry News
    Stratnews GlobalStratnews Global
    Home » Huawei Launches Safer DeepSeek-R1-Safe AI Model with Stronger Security Controls

    Huawei Launches Safer DeepSeek-R1-Safe AI Model with Stronger Security Controls

    ReshamBy ReshamSeptember 25, 2025Updated:September 25, 2025 AI and Robotics No Comments2 Mins Read
    Huawei Deepseek

    Huawei Unveils Safer AI Model DeepSeek-R1-Safe

    Chinese technology company Huawei has introduced a safety-focused version of the DeepSeek artificial intelligence model, claiming it is almost entirely effective in blocking politically sensitive discussions.

    Focus on Security and Compliance

    Chinese regulators require all domestic AI models to align with the country’s socialist values before public release. These rules ensure strict compliance with national speech regulations. Huawei announced on its WeChat account that it trained the model using 1,000 of its Ascend AI chips. The model, named DeepSeek-R1-Safe, is based on the open-source DeepSeek-R1 but adjusted for stronger safeguards.

    Huawei partnered with Zhejiang University, which is also the alma mater of DeepSeek’s founder Liang Wenfeng. However, both DeepSeek and Liang were not directly involved in this project. DeepSeek has yet to comment on Huawei’s announcement.

    Industry Adoption of DeepSeek

    The release of DeepSeek’s advanced models, R1 and V3, surprised Silicon Valley and global investors, leading to a decline in Western AI stock values earlier this year. Since then, Chinese industries and organisations have actively modified and deployed DeepSeek models.

    AI chatbots in China, such as Baidu’s Ernie Bot, already avoid questions on domestic politics or other sensitive subjects. Huawei’s new model continues this trend, focusing on reinforcing these restrictions.

    Testing and Performance

    Huawei stated that DeepSeek-R1-Safe achieved nearly 100% success in blocking harmful or politically sensitive content, as well as preventing incitement to illegal acts. However, its effectiveness fell to 40% in tests involving role-playing scenarios, encrypted codes, or disguised prompts.

    Overall, the model scored an 83% success rate in comprehensive security testing. This performance was 8% to 15% higher than comparable AI models, such as Alibaba’s Qwen-235B and DeepSeek-R1-671B. Importantly, DeepSeek-R1-Safe demonstrated less than 1% performance loss compared to the original DeepSeek-R1, according to Huawei.

    The announcement coincides with Huawei’s annual Connect conference in Shanghai, where the company also revealed new chip and computing power roadmaps.

    with inputs from Reuters

    Author

    • Resham
      Resham

      Research Associate at Interstellar.| China Scholar | China Social Media & Foreign Affairs|

      View all posts
    Featured
    Share. Facebook Twitter Pinterest LinkedIn Tumblr Email Reddit Telegram WhatsApp
    Resham
    • X (Twitter)

    Research Associate at Interstellar.| China Scholar | China Social Media & Foreign Affairs|

    Keep Reading

    RWE pulls out of Namibia’s $10bn Hyphen green ammonia venture

    Euclid Consortium Unveils Largest-Ever Universe Simulation with 3.4 Billion Galaxies

    UN Arms Embargo Returns as Iran Warns of Retaliation

    China’s Chang’e-6 Mission Wins IAF World Space Award at Sydney Congress

    China’s Giant Antenna Powers Global Race for Lithium and Rare Earths

    BRICS and IBSA Advance Space and Technology Cooperation amid Global Realignments

    Add A Comment
    Leave A Reply Cancel Reply

    Latest Posts

    RWE pulls out of Namibia’s $10bn Hyphen green ammonia venture

    September 29, 2025

    Euclid Consortium Unveils Largest-Ever Universe Simulation with 3.4 Billion Galaxies

    September 29, 2025

    UN Arms Embargo Returns as Iran Warns of Retaliation

    September 29, 2025

    China’s Chang’e-6 Mission Wins IAF World Space Award at Sydney Congress

    September 29, 2025

    China’s Giant Antenna Powers Global Race for Lithium and Rare Earths

    September 29, 2025

    BRICS and IBSA Advance Space and Technology Cooperation amid Global Realignments

    September 27, 2025

    Judge Gives Preliminary Approval to $1.5B Anthropic Copyright Deal

    September 26, 2025

    Microsoft Blocks Israeli Military Use of Azure After Surveillance Findings

    September 26, 2025

    Trump Signs Order Approving $14 Billion TikTok U.S. Deal

    September 26, 2025

    MeitY Begins Preparatory Phase of Special Campaign 5.0 to Streamline Records

    September 26, 2025

    Subscribe to News

    Get the latest sports news from NewsSite about world, sports and politics.

    • Astronomical Events
    • Space Missions
    • Industry News
    • Science
    StratNewsGlobal Tech
    Facebook X (Twitter) Instagram LinkedIn YouTube
    © 2025 StratNews Global, A unit of BharatShakti Communications LLP
    • About Us
    • Contributors
    • Copyright
    • Contact
    • Write for Us

    Type above and press Enter to search. Press Esc to cancel.

    ×