Anthropic, di company wey dey behind Claude chatbot, don suspend dema new powerful AI models Fable 5 and Mythos 5 just few days after release. Dis one na follow US government directive wey cite national security concerns.
According to Anthropic statement, di US government issue export control order wey say make dem block any foreign national from accessing di models. Di directive come from Commerce Secretary Howard Lutnick and Bureau of Industry and Security officials.
Anthropic tok say dem no get choice but to disable di models for all customers, no be only foreign nationals. Dis na first time federal government don force AI company to take public model offline.
Di government identify one technique wey fit allow users to bypass some safeguards for Fable 5. Anthropic say dem only get verbal evidence of dis issue and di same weakness likely dey for other AI systems.
Di company disagree with di decision, saying di finding of narrow potential jailbreak no be reason to recall model wey hundreds of millions dey use. Anthropic release Fable 5 and Mythos 5 on Tuesday, calling dem di most powerful AI systems dem ever share.
Fable 5 get strong restrictions on questions about cybersecurity and biology. Mythos 5 na for trusted partners like cybersecurity companies. Anthropic believe di directive come from misunderstanding and hope to restore access soon.
Dis matter na latest for di wahala between Anthropic and Trump administration. Earlier for February, President Trump and Defense Secretary Pete Hegseth bar Anthropic products from federal agencies after di company seek stronger guardrails for Pentagon use. Trump call Anthropic leftwing nut jobs for Truth Social.
Anthropic sue di administration, arguing di order violate freedom of speech. Despite di dispute, dem don seem to make amends recently. Last week, National Security Agency reportedly dey use Mythos for offensive cyberattacks.
On June 2, Trump sign executive order on AI wey direct agencies to bolster cyber defenses and design mechanism for early access to powerful AI models. Anthropic say government suppose get power to block unsafe deployments through transparent statutory process, but dis action no follow those principles.
Gina Neff, Professor of Responsible AI for Queen Mary University London, warn say restricting access fit limit development and safe testing of AI systems, and fit restrict collaboration with governments worldwide. She say we dey for uncharted territory as AI tools dey improve rapidly.
UK government AI Security Institute test find say di model fit exploit defenses 73% of di time. Anthropic self tok say Fable 5 capabilities exceed any model dem ever release, and without safeguards, di power for cybersecurity fit cause serious damage.
