CrowdStrike CEO George Kurtz on AI safety concerns: This is a solvable problem

phone and call now to see what your case could be worth. >> CRWD CROWDSTRIKE ON A TOTAL TEAR, HITTING A NEW ALL TIME HIGH YESTERDAY. IT LOOKS LIKE IT'S GOING TO HAVE A NEW ONE TODAY, UP NEARLY 180% IN THE PAST SIX MONTHS. NOT BAD. COMPANY CONTINUING TO BENEFIT FROM AI DRIVEN CYBERSECURITY DEMAND. JOINING US NOW. GREAT HONOR POST NINE. GEORGE KURTZ CROWDSTRIKE CEO GEORGE, FIRST OF ALL, THANK YOU FOR SHOWING UP BECAUSE I'VE GOT TO TELL YOU, WE'RE GOING TO PUT WE'RE GOING TO PUT TO BED SOME STORIES TODAY. WE'RE GOING TO BE ABLE TO MAKE IT SO IT'S NOT STEPHEN KING, IT'S NOT THE STAND. IT'S AN ENGINEERING PROBLEM AND IT CAN BE BEATEN. I'M TALKING ABOUT THE AI CRAZINESS THAT WE HEAR FROM A COUPLE OF COMPANIES. WHERE ARE YOU ON THINGS SAY LIKE THE NVIDIA OPEN AGENT SAFETY PLATFORM, WHICH MADE ME FEEL PRETTY GOOD. AND I THOUGHT THAT IT BECAME AN ENGINEERING PROBLEM WITH THAT. >> WELL. >> WE'RE OBVIOUSLY PART OF THAT. WE WORK VERY CLOSELY WITH NVIDIA, AND WHEN WE THINK ABOUT WHAT AN AGENT DOES, AN AGENT IS A PROGRAM THAT ACTUALLY RUNS AND TOUCHES OTHER FILES. IT TOUCHES DATA FILES, IT TAKES ACTION. AND WHAT'S IMPORTANT IS THE MODEL PROVIDERS ARE FOCUSED ON TRYING TO BUILD GUARDRAILS AROUND HOW THE MODEL ACTUALLY THINKS. BUT WHEN YOU PUT IT IN PRODUCTION, YOU NEED TO SEE WHAT IT ACTUALLY DOES. YOU NEED TO UNDERSTAND ITS ACTIONS. AND ONE OF THE THINGS THAT WE FOCUSED ON IS BUILDING THE FLIGHT RECORDER FOR WHAT THESE AGENTS ARE DOING. YOU HAVE TO HAVE TRACEABILITY ON WHAT THEY'RE ACTUALLY DOING, AND YOU HAVE TO TIE THAT INTO THE ABILITY TO ENFORCE IT, WHICH IS WHAT WE'VE COLLABORATED ON WITH NVIDIA. SO THIS IS A SOLVABLE PROBLEM. YOU JUST HAVE TO HAVE THE RIGHT PARTNERSHIP AND TECHNOLOGIES COMING TOGETHER TO BUILD SAFETY FOR COMPANIES THAT ARE USING THESE AGENTS. >> OKAY, SO YOU'VE GOT PEOPLE ON ONE SIDE WHO SAY IT'S A SOLVABLE PROBLEM, AND THEN PEOPLE ON THE OTHER SIDE WHO SAY, WE GOT TO SLOW THINGS DOWN BECAUSE WE HAVE NO IDEA WHAT CAN HAPPEN. WHY ARE THE PEOPLE IN THE LATTER CAMP SO ASCENDANT RIGHT NOW IN THE DISCOURSE OF THE GENERAL POPULACE, NOT WHERE YOU ARE RIGHT NOW, BUT THE GEN POP, PARTICULARLY AHEAD OF AN ELECTION? >> WELL, LOOK, I THINK THERE'S A LOT OF DIFFERENT MOTIVATIONS OUT THERE OF WHO WANTS TO DO WHAT AND SLOW THINGS DOWN. I THINK WHEN YOU LOOK AT AI IN GENERAL, IT'S BEEN A FOUNDATIONAL TECHNOLOGY. I MEAN, IT'S A IT'S THE GREATEST SEA CHANGE THAT I'VE SEEN IN MY IN MY LIFETIME AND CERTAINLY MY TECHNOLOGY CAREER. SO YOU'VE GOT TO BE ABLE TO HARNESS IT. THIS ISN'T IN RACING. YOU DON'T SLOW THINGS DOWN. YOU WANT TO HAVE THE GREATEST AMOUNT OF SAFETY SO THAT YOU CAN GO FASTER. AND I THINK THIS IS AN IMPORTANT TOPIC. OBVIOUSLY, SAFETY AND SECURITY IS IMPORTANT, BUT YOU GOT TO DO IT IN THE RIGHT WAY. AND IT DOESN'T MEAN THAT YOU HAVE TO BASICALLY SLOW EVERYTHING DOWN AND THEN GIVE OUR ADVERSARIES, LIKE CHINA AND OTHERS, AN ADVANTAGE AHEAD OF US. SO I THINK THIS IS WHY FOLKS ARE COMING TOGETHER ON ONE SIDE VERSUS THE OTHER SAYING IT'S SOLVABLE. WE JUST HAVE TO DO IT IN THE RIGHT WAY, AND WE'VE GOT TO DO IT WITHOUT A MASSIVE AMOUNT OF REGULATION. >> WELL, THEN WHAT IS IT? WHAT'S SOLVABLE? I WANT TO TRY AND UNDERSTAND THIS BECAUSE YOU HEAR DIFFERENT THINGS FROM DIFFERENT PEOPLE. BILL GATES DID AN INTERVIEW I LISTENED TO LAST WEEK TALKING ABOUT, WE NEED A SUPERVISORY LEVEL ACROSS THE WORLD, GIVEN THE OPEN MODELS, NOT JUST NOT JUST THE CLOSED MODELS THAT WE ALL TALK ABOUT HERE. LIKE, SO WHAT ACTUALLY WOULD WORK? >> WELL, WHAT ACTUALLY WORKS IS LET'S START WITH THE RIGHT STANDARDS, RIGHT? SO IF YOU LOOK AT WHAT WHAT'S WHAT THE ORGANIZATIONS ARE FOCUSED ON NOW IS SELF-REGULATION. BUT THINGS LIKE THE NIST STANDARDS, WHICH COME OUT OF THE US GOVERNMENT, RIGHT? THESE ARE THEY'RE WORKING ON UPDATED ONES FOR AI, WHICH BASICALLY IS FOCUSED ON HOW DO WE BUILD AND TRAIN MODELS AND PUT GUARDRAILS AROUND THEM. AND THEN YOU HAVE BOARD OVERSIGHT, YOU'VE GOT EXTERNAL AUDITORS. THOSE ARE ALL GREAT STEPS. FANTASTIC. BUT THEN WHEN YOU PUT IT IN PRODUCTION, YOU HAVE TO SEE WHAT IT ACTUALLY DOES. THINK ABOUT A PLANE. YOU BUILD A PLANE, THERE'S CERTAIN STANDARDS, THEN YOU PUT IT INTO PRODUCTION, YOU FLY IT, AND WHEN SOMETHING HAPPENS, YOU HAVE TO GO BACK TO THE BLACK BOX. WHAT DID IT DO? WHAT REALLY HAPPENED? SO THERE'S RUNTIME CONTROL. >> YOU'VE ALSO GOT A FEDERAL REGULATOR CALLED THE FAA THAT'S LOOKING OVER YOUR SHOULDER ALL THE TIME. >> YOU DO. BUT I THINK AT THE END OF THE DAY, YOU KNOW, WE CAN WAIT FOR THE NEXT FIVE YEARS TO GET AI MODELS OUT OR WE CAN BE COMPETITIVE AND, AND HAVE SELF-POLICING. >> YEAH. BUT WE DON'T WANT THE PLANE TO CRASH. GEORGE. >> THAT'S WHY YOU'VE GOT REGULATION AROUND THOSE PLANES BECAUSE YOU'VE GOT PEOPLE INVOLVED IN, IN ACTUALLY CREATING THEM. BUT WHAT FROM A PLANE MANUFACTURER PERSPECTIVE, IT TAKES YEARS TO GET SOMETHING CERTIFIED. DO WE WANT TO WAIT FIVE YEARS TO HAVE A MODEL CERTIFIED? >> WELL, YES, BECAUSE IN FIVE YEARS THERE ARE PEOPLE WHO HAVE BEEN IN IN THE SCRUM WHO SAY THAT WE COULD ALL BE DEAD, WHICH IS A SUBOPTIMAL SITUATION. >> GEORGE. NONSENSE, JIM. WE'RE NOT GOING TO BE DEAD IN 2030 WITH A 10% CHANCE. >> WELL, THEN WHY SAY IT? WHY DO YOU THINK JUST PEOPLE JUST SAY IT? BECAUSE IT'S LIKE THEY'RE NOVELISTS. >> I THINK PEOPLE HAVE THEIR OWN MOTIVATIONS FOR WHY THEY WANT TO SAY IT. >> I KNOW, BUT SOME OF THEM ARE VERY CLOSE TO THE TECHNOLOGY, GEORGE. I MEAN, THEY'RE WORKING IN THE LABS AND THE SAFETY RESEARCHERS WHO ARE ACTUALLY LOOKING AT ALIGNMENT AND THEY'RE SAYING, WHOA, I DON'T LIKE WHAT I'M SEEING. >> WELL, I THINK THEY SHOULD CONSULT THE SECURITY GUYS MORE OFTEN BECAUSE IF YOU LOOK AT WHAT AN AGENT DOES, IT'S A PROGRAM THAT RUNS JUST NON-DETERMINISTICALLY. AND THESE ARE THINGS THAT CROWDSTRIKE HAS BEEN FOCUSED ON PROTECTING WHAT A HUMAN DOES FOR MANY, MANY YEARS, RIGHT? AN AGENT IS SUPERHUMAN, TOUCHES A LOT OF DIFFERENT FILES, BUT THE TECHNOLOGY IS THERE. AND I THINK WHAT'S IMPORTANT TO REMEMBER IS THAT THE SECURITY INDUSTRY NOW IS MOVING FROM KEEPING ALL THE BAD GUYS OUT TO ENABLING AI TO BE USED SAFELY. SO WE MOVE FROM SECURITY TO SAFETY. AND THAT'S NOT JUST A BIGGER MARKET. IT'S A DIFFERENT MARKET. IT'S ALLOWING THE CEO TO EXPAND THEIR BUSINESS AND RUN FASTER. AND THIS IS THE OPPORTUNITY THAT WE SEE IN FRONT OF US, NOT ONLY FOR THE SECURITY MARKET, BUT FOR CROWDSTRIKE. >> MEANTIME, YOU'RE ANNOUNCING, I THINK IT'S YOUR FOURTH STARTUP ACCELERATOR WITH AWS AND NVIDIA. YES. SO TELL US ABOUT THAT. WHAT DO APPLICANTS DO? >> WELL, WE BELIEVE IN THE ECOSYSTEM AND WORKING WITH SMALLER COMPANIES. THERE'S LOTS OF COMPANIES AROUND THE WORLD THAT ARE INCREDIBLE AND WE WANT TO BE ABLE TO INCUBATE THEM. WE WANT TO MAKE SURE THEY CAN LEVERAGE A PLATFORM LIKE CROWDSTRIKE. FALCON WANT TO LEVERAGE TECHNOLOGIES FROM A FOUNDATIONAL LEVEL LIKE NVIDIA, WANT TO LEVERAGE THE HYPERSCALERS, AWS. SO WE'RE PUTTING THEM ALL TOGETHER AND WE'RE BASICALLY LOOKING FOR THE NEXT GEMS, THE NEXT LEADERS, WHO'S THE NEXT POTENTIAL CROWDSTRIKE. AND WE WANT TO MAKE SURE THAT THEY'VE GOT EVERYTHING THAT THAT THEY CAN TO BE SUCCESSFUL. >> WELL, GEORGE, TODAY THIS IS GENTLEMEN, THIS WEEKEND WITH OVER 100 INDUSTRY PARTNERS, WE INTRODUCED NVIDIA OPEN AGENT SAFETY PLATFORM. ANTHROPIC WAS IN WHERE WAS OPEN AI. DOES IT MATTER THAT A COMPANY LIKE OPENAI DOES NOT JOIN THIS? >> LOOK, I THINK IT'S A GOOD QUESTION FOR OPEN AI. AT THE END OF THE DAY, THEY'RE ALL INCENTED. AND I THINK THE MARKET IS GOING TO BASICALLY DETERMINE WHO'S GOT THE GREATEST SAFETY AROUND THEIR MODELS, WHO'S DELIVERING THE RIGHT ENTERPRISE FEATURES THAT THEY NEED, WHO'S ALLOWING DATA SOVEREIGNTY TO TAKE PLACE? THERE'S A LOT OF CONCERN ABOUT WHERE ALL THIS DATA GOES AND WHERE WHAT HAPPENS WHEN IT HITS THE MODEL PROVIDERS. AND I THINK ULTIMATELY, THE MARKET IS GOING TO DETERMINE THAT IF YOU'VE GOT SAFE MODELS, YOU'VE GOT GOOD MODELS, PEOPLE ARE GOING TO PAY FOR IT. AND I THINK THAT'S WHAT'S WHAT'S IMPORTANT. >> CAN I COME BACK TO OUR WORLD FOR ONE SECOND HERE, THE FINANCIAL WORLD IN GENERAL? I MEAN, WHAT I HEAR, CERTAINLY FROM LEADERS OF OF FINANCIAL SERVICES FIRMS AS WELL, BOTH LARGER AND SMALLER IS THE MAIN FEAR IS THE CYBER HACK OF SOME KIND, NEVER ON THEM, BUT ON ONE OF THEIR COMPETITORS. AND WHAT THAT COULD DO AND THE CHAOS THAT COULD UNLEASH, HOW REAL IS THAT THREAT? >> WELL, I THINK FROM THE STANDPOINT OF CYBER HACKS, IT'S A THREAT FOR EVERYONE, RIGHT? AND WHETHER IT'S A MODEL PROVIDER BEING HACKED OR WHETHER IT'S SOMEONE GETTING HACKED FROM THEM, OBVIOUSLY THERE ARE THINGS THAT WE HAVE TO ACCOUNT FOR AND PROTECT AGAINST, AND THAT'S WHY IT'S SO IMPORTANT TO UNDERSTAND WHAT THE AGENTS ARE DOING AND WHAT THEY'RE TOUCHING AND WHERE THEY'RE GOING. I MEAN, IF YOU IF YOU LOOK AT ONE OF THE HACKS IN AUSTRALIA, IT WAS 84 DAYS BEFORE ANYBODY KNEW WHAT HAPPENED, RIGHT. WHICH IS WHY WHAT I TALKED ABOUT THIS BLACK BOX TRACEABILITY IS SO IMPORTANT. IF YOU'RE ACTUALLY LOOKING AND YOU UNDERSTAND WHAT THE AGENT IS DOING, IT'S EASY TO TELL WHAT IS HAPPENING. THERE'S A LOT OF INFORMATION, BUT YOU HAVE TO BE ABLE TO SORT THROUGH IT. AND THESE ARE THINGS THAT WE'RE FOCUSED ON. >> TODD MCKINNON AT OKTA SAYS THERE IS AN INDIVIDUAL IDENTIFIER OF EVERY SINGLE ONE OF THESE, AND YOU CAN FIND OUT WHERE THEY ARE AND YOU CAN STOP THEM WITH YOU. WHY? WHY IS THIS NOT JUST CONSIDERED AN ENGINEERING ISSUE? WHY IS IT CONSIDERED SOME SORT OF ETHEREUM, ALMOST RELIGIOUS DEATH THAT WE ALL FACE? >> I MEAN, WHAT TODD'S TALKING ABOUT IS IDENTITY, WHICH IS INCREDIBLY IMPORTANT FOR HIS BUSINESS AND OUR BUSINESS AS WELL. AND FROM AN AGENT PERSPECTIVE, YOU WANT TO GIVE IT THE LEAST AMOUNT OF PRIVILEGES TO ONLY DO WHAT IT NEEDS TO DO. THE PROBLEM IS IF WE THINK ABOUT JUST THINK ABOUT AN EMPLOYEE, YOU WOULD NEVER ONBOARD AN EMPLOYEE, GIVE THEM SUPERUSER PRIVILEGES AND NEVER DO A BACKGROUND CHECK. THAT'S WHAT YOU DO WITH AN AGENT. YOU TURN IT ON, YOU GIVE IT PRIVILEGES. >> NEVER. WHAT HAPPENED? WHAT HAPPENED WITH THAT FLIGHT? THE FLIGHT YOU BOUGHT? I MEAN, IF YOU WANT TO USE THAT ANALOGY, YOU NEVER GIVE IT A BACKGROUND CHECK. WELL, THEY DIDN'T THEY DIDN'T GIVE IT A BACKGROUND CHECK. THEY DIDN'T GIVE THAT PILOT A BACKGROUND CHECK. >> AND THIS IS WHAT WE'RE DEALING WITH. SO I MEAN, RIGHT NOW, WHAT ORGANIZATIONS ARE FOCUSED ON IS FINDING THE ROGUE AI IN THEIR ENVIRONMENTS BECAUSE IT'S EVERYWHERE. AND TWO, HOW DO THEY UNDERSTAND WHAT IT'S ACTUALLY DOING AND PUT GUARDRAILS AT RUNTIME? OKAY. REMEMBER THERE'S TRAINING THE MODEL, THE MODEL HAS EXPECTED BEHAVIOR, BUT JUST LIKE ANYTHING ELSE, IT MAY DO SOMETHING UNEXPECTED. YES. RIGHT. IT'S NON-DETERMINISTIC. THIS IS THE PROBLEM. THAT'S, THAT'S, THAT'S THE ISSUE THAT THE LABS ARE DEALING WITH. WHEN YOU PUT IT IN PRODUCTION, YOU HAVE TO LOOK AT WHAT IT'S ACTUALLY DOING AND DETERMINE THE PROP THAT YOU PUT IN. DID IT ACTUALLY TAKE THE ACTIONS YOU THOUGHT IT WAS GOING TO TAKE? AND THAT'S WHAT WE'RE FOCUSED ON. THAT'S WHY CROWDSTRIKE HAS SEEN SUCH TAILWINDS, BECAUSE EVERY COMPANY ON THE PLANET IS SAYING, WHAT IS THIS THING DOING WHEN SOMETHING BAD HAPPENS? DO YOU THINK THE FINANCIAL SERVICES COMPANIES ALL RIGHT AROUND US HERE IN WALL STREET ARE GOING