US Export Ban on Anthropic Models Leaves India Builders Exposed
An Amazon security paper and a quiet word from Andy Jassy pushed Washington to pull Anthropic's Fable 5 and Mythos 5 from foreign users. India's AI builders just lost their safety net.
The News
The United States government has ordered Anthropic to switch off two of its most capable models, Fable 5 and Mythos 5, for everyone outside the country. The move was first reported by the Wall Street Journal and picked up by The Verge, and it followed both a piece of cybersecurity research from Amazon and a run of private conversations between Amazon chief executive Andy Jassy and the Trump administration.
Amazon's researchers reportedly found that Fable 5 could be steered, through a carefully ordered sequence of prompts, into producing material useful for cyberattacks. The work is said to have surfaced security flaws across at least four separate software programmes when the model was given the right queries.
Once Jassy flagged the findings with senior officials, the White House gathered to shape a response. National Cyber Director Sean Cairncross and Commerce Secretary Howard Lutnick were among those drawn in, and President Trump signed off on an export-control measure that blocks the two models from foreign governments, companies and individuals. Anthropic has since apologised for the disruption to its service.
Why It Matters
The detail that makes this remarkable is the source of the pressure. Amazon has committed billions of dollars to Anthropic, so a clampdown driven partly by its own backer's research is close to unprecedented in the current AI cycle.
It also marks a shift in how Washington thinks about this technology. For years the export-control conversation centred on hardware, from the curbs on advanced Nvidia chips bound for China in 2022 and 2023 to the licensing rules that followed. Treating a frontier model's reasoning ability as itself a controlled export pushes that logic up the stack, from silicon to software.
Anthropic has pushed back hard, arguing that holding every model to this bar would in practice freeze new launches across the sector, and describing the order as "a likely misunderstanding". The standoff lands at an awkward moment, with the company widely reported to be preparing for a possible public listing.
Indian Angle
The phrase doing the heavy lifting here is "foreign users", and that bracket includes India. A large number of Indian startups, fintechs and enterprise teams have built products on Claude, many of them through Amazon's own Bedrock platform on AWS. An abrupt suspension of the strongest models leaves those teams scrambling for fallbacks, retuning prompts and explaining outages to their own customers.
It also sharpens the case for home-grown capability. Sovereign efforts such as Sarvam and Ola-backed Krutrim, along with the government's IndiaAI Mission, have long argued that depending on a handful of American labs is a strategic risk. An overnight policy decision in Washington that strands Indian developers is exactly the scenario they have warned about.
For policymakers at MeitY, the episode is a live case study in supply concentration. For founders, the lesson is more practical: model access is now a geopolitical variable, and a serious product cannot rest on a single provider that a foreign government can switch off.
FAQ
Which models are affected?
The directive covers Anthropic's Fable 5 and Mythos 5, two of its most advanced systems. Access has been cut for users outside the United States, including foreign governments, companies and individuals, while domestic access continues for now.
Why did Amazon's research matter so much?
Amazon reported that Fable 5 could be prompted into revealing information useful for cyberattacks, including flaws in at least four software programmes. Coming from a major Anthropic investor, the findings carried unusual weight in Washington.
What has Anthropic said?
The company apologised for the service interruption and pushed back on the reasoning, warning that the same standard applied industry-wide would stall new model releases. It described the order as "a likely misunderstanding".
What should Indian developers do now?
Teams relying on Claude, especially via AWS Bedrock, should test fallback models, isolate critical workflows from any single provider, and watch for guidance from cloud partners on which regions and models remain available.
This story was reported by The Verge. Read the full original coverage at The Verge.