TZUTC: -0500   
   MSGID: 1497.consprcy@1:2320/105 2d29ea63   
   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   
   'The models are really devious': Sam Altman's hardware chief says OpenAI    
   wants kill switches built into hardware in case things go wrong   
      
   Date:   
   Fri, 12 Sep 2025 20:46:00 +0000   
      
   Description:   
   OpenAI hardware leader warns that devious AI models require real time    
   hardware kill switches and deeper safeguards built directly into chips.   
      
   FULL STORY   
      
   A senior OpenAI executive has warned that future AI infrastructure will   
   require hardware-level safety features, including kill switches.    
      
   Richard Ho, head of hardware at the company, made the remarks during his   
   keynote at the AI Infra Summit in Santa Clara.    
      
   It has to be built into the hardware, Ho said. Today a lot of safety work is   
   in the software. It assumes that your hardware is secure. It assumes that    
   your hardware will do the right thing. It assumes that you can pull the plug   
   on the hardware. I am not saying that we cant pull the plug on that hardware,   
   but I am telling you that these things are devious, the models are really   
   devious, and so as a hardware guy, I want to make sure of that.   
      
   Silicon-level safety measures   
      
   Ho argued that the growth of generative AI is forcing a rethink of system   
   architecture and described how future agents will be long-lived, interacting   
   in the background even when a user is not actively engaged.    
      
   This shift requires memory-rich, low-latency infrastructure to manage   
   continuous sessions and communication across multiple agents.    
      
   Networking, Ho said, is becoming a bottleneck. Were going to have to have   
   real-time tools in these  meaning that these agents communicate with each   
   other. Some of them might be looking at a tool, some might be doing a website   
   search. Others are thinking, and others need to talk to each other.    
      
   Ho outlined several hardware challenges that must be addressed, including   
   limits on high-bandwidth memory, the need for 2.5D and 3D chip integration,   
   advances in optics, and extreme power requirements that could reach 1    
   megawatt per rack.    
      
   The safety measures OpenAI put forward include real-time kill switches built   
   into AI clusters, telemetry to detect signs of abnormal behavior, and secure   
   execution paths in CPUs and accelerators.    
      
   Ho wrapped things up by saying, We dont have good benchmarks for agent-aware   
   architectures and hardware, and I think it is important to know about latency   
   walls and latency tails, what is the efficiency and power and things like   
   that. We need to have good observability as a hardware feature, not just as a   
   debug tool, but built in and constantly monitoring our hardware."    
      
   "Networking is a real important thing, and as we head towards optical, it is   
   unclear that the reliability of the network is there today. We need to get   
   there with enough testing of these optical testbeds and these other   
   communication testbeds that show that we actually have the reliability.    
      
   Via The Next Platform   
      
   ======================================================================   
   Link to news story:   
   https://www.techradar.com/ai-platforms-assistants/chatgpt/the-models-are-reall   
   y-devious-sam-altmans-hardware-chief-says-openai-wants-kill-switches-built-int   
   o-hardware-in-case-things-go-wrong   
      
   $$   
   --- 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: 226/30 227/114 229/110 111 206 300 307 317 400 426 428 470   
   SEEN-BY: 229/664 700 705 266/512 291/111 320/219 322/757 342/200 396/45   
   SEEN-BY: 460/58 712/848 902/26 2320/0 105 304 3634/12 5075/35   
   PATH: 2320/105 229/426