TZUTC: -0500   
   MSGID: 1750.consprcy@1:2320/105 2d8c5049   
   PID: Synchronet 3.21a-Linux master/123f2d28a Jul 12 2025 GCC 12.2.0   
   TID: SBBSecho 3.28-Linux master/123f2d28a Jul 12 2025 GCC 12.2.0   
   BBSID: CAPCITY2   
   CHRS: ASCII 1   
   FORMAT: flowed   
   DeepSeek took off as an AI superstar a year ago - but could it also be a    
   major security risk? These experts think so   
      
   Date:   
   Tue, 25 Nov 2025 20:28:00 +0000   
      
   Description:   
   DeepSeek-R1s code output becomes insecure when political topics are included,   
   revealing hidden censorship and serious risks for enterprise deployments.   
      
   FULL STORY   
      
   When it released in January 2025, DeepSeek-R1, a Chinese large language model   
   ( LLM ) caused a frenzy and has since been widely adopted as a coding   
   assistant.    
      
   However, independent tests by CrowdStrike claim the models output can vary   
   significantly depending on seemingly irrelevant contextual modifiers.    
      
   The team tested 50 coding tasks across multiple security categories with 121   
   trigger-word configurations, with each prompt run five times, totaling 30,250   
   tests, and the responses were evaluated using a vulnerability score from 1   
   (secure) to 5 (critically vulnerable).   
      
   Politically sensitive topics corrupt output   
      
   The report reveals that when political or sensitive terms such as Falun Gong,   
   Uyghurs, or Tibet were included in prompts, DeepSeek-R1 produced code with   
   serious security vulnerabilities. These included hard-coded secrets, insecure   
   handling of user input, and in some cases, completely invalid code.   
      
   The researchers claim these politically sensitive triggers can increase the   
   likelihood of insecure output by 50% compared to baseline prompts without    
   such words.    
      
   In experiments involving more complex prompts, DeepSeek-R1 produced    
   functional applications with signup forms, databases, and admin panels.    
   However, these applications lacked basic session management and   
   authentication, leaving sensitive user data exposed - and across repeated   
   trials, up to 35% of implementations included weak or absent password    
   hashing.    
      
   Simpler prompts, such as requests for football fan club websites, produced   
   fewer severe issues.    
      
   CrowdStrike, therefore, claims that politically sensitive triggers   
   disproportionately impacted code security. The model also demonstrated an   
   intrinsic kill switch - as in nearly half of the cases, DeepSeek-R1 refused to   
   generate code for certain politically sensitive prompts after initially   
   planning a response.  Examination of the reasoning traces showed the model   
   internally produced a technical plan but ultimately declined assistance.   
      
   The researchers believe this reflects censorship built into the model to   
   comply with Chinese regulations, and noted the models political and ethical   
   alignment can directly affect the reliability of the generated code.    
   For politically sensitive topics, LLMs generally tend to give the ideas of   
   mainstream media, but this could be in stark contrast with other reliable    
   news outlets.    
      
   DeepSeek-R1 remains a capable coding model, but these experiments show that    
   AI tools , including ChatGPT and others, can introduce hidden risks in   
   enterprise environments.  Organizations relying on LLM-generated code should   
   perform thorough internal testing before deployment.  Also, security layers   
   such as a firewall and antivirus remain essential, as the model may produce   
   unpredictable or vulnerable outputs.   
      
   Biases baked into the model weights create a novel supply-chain risk that   
   could affect code quality and overall system security.    
      
   ======================================================================   
   Link to news story:   
   https://www.techradar.com/pro/deepseek-took-off-as-an-ai-superstar-a-year-ago-   
   but-could-it-also-be-a-major-security-risk-these-experts-think-so   
      
   $$   
   --- SBBSecho 3.28-Linux   
    * Origin: capitolcityonline.net * Telnet/SSH:2022/HTTP (1:2320/105)   
   SEEN-BY: 105/81 106/201 128/187 129/14 305 153/7715 154/110 218/700   
   SEEN-BY: 226/30 227/114 229/110 206 300 307 317 400 426 428 470 664   
   SEEN-BY: 229/700 705 266/512 291/111 320/219 322/757 342/200 396/45   
   SEEN-BY: 460/58 633/280 712/848 902/26 2320/0 105 304 3634/12 5075/35   
   PATH: 2320/105 229/426