Topics
recent
AI
Amazon
Image Credits:Kimberly White/Getty Images for TechCrunch
Apps
Biotech & Health
Climate
Image Credits:Kimberly White/Getty Images for TechCrunch
Cloud Computing
Commerce
Crypto
endeavour
EVs
Fintech
Fundraising
Gadgets
punt
Government & Policy
Hardware
Layoffs
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
Security
societal
Space
inauguration
TikTok
Transportation
speculation
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
meet Us
Anthropic ’s CEO Dario Amodei is distressed about competitor DeepSeek , the Chinese AI companythat took Silicon Valley by stormwithits R1 model . And his concerns could be more serious than the distinctive I raised about DeepSeek sending drug user data back to China .
Inan interviewon Jordan Schneider ’s ChinaTalk podcast , Amodei said DeepSeek yield rarified information about bioweapons in a rubber trial run run by Anthropic .
DeepSeek ’s carrying into action was “ the worst of basically any exemplar we ’d ever test , ” Amodei lay claim . “ It had absolutely no occlusion whatsoever against generating this information . ”
Amodei stated that this was part of evaluation Anthropic routinely runs on various AI models to assess their potential national security department risks . His team looks at whether fashion model can beget bioweapon - related information that is n’t well found on Google or in textbooks . anthropical positions itself as the AI foundational framework providerthat take prophylactic seriously .
Amodei said he did n’t think DeepSeek ’s models today are “ literally serious ” in provide rarified and serious information but that they might be in the near future . Although he praise DeepSeek ’s team as “ gifted engineer , ” he advise the company to “ take badly these AI safety considerations . ”
Amodei has alsosupported strong export controlson chips to China , citing headache that they could give China ’s war machine an edge .
Amodei did n’t clarify in the ChinaTalk interview which DeepSeek model Anthropic tested , nor did he give more proficient details about these trial run . Anthropic did n’t immediately reply to a postulation for scuttlebutt from TechCrunch . Neither did DeepSeek .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
DeepSeek ’s rise has set off concerns about its safety elsewhere , too . For exemplar , Cisco certificate researcherssaid last weekthat DeepSeek R1 failed to forget any harmful prompt in its guard tests , achieving a 100 % jailbreak success rate .
Cisco did n’t cite bioweapons but said it was able to get DeepSeek to generate harmful information about cybercrime and other illegal activities . It ’s worth mentioning , though , that Meta ’s Llama-3.1 - 405B and OpenAI ’s GPT-4o also had gamy failure rates of 96 % and 86 % , severally .
It remains to be seen whether refuge concerns like these will make a serious nick in DeepSeek’srapid acceptance . company likeAWSandMicrosofthave publicly touted integrate R1 into their cloud platform — ironically enough , make that Amazon isAnthropic ’s biggest investor .
On the other manus , there ’s a raise listof countries , caller , and specially government formation like the U.S. Navy andthe Pentagonthat have initiate banning DeepSeek .
metre will tell if these efforts get on or if DeepSeek ’s global hike will carry on . Either direction , Amodei tell he does look at DeepSeek a new rival that ’s on the tier of the U.S. ’s top AI companies .
“ The fresh fact here is that there ’s a novel competitor , ” he said on ChinaTalk . “ In the big caller that can prepare AI — Anthropic , OpenAI , Google , perhaps Meta and xAI — now DeepSeek is maybe being added to that class . ”
