Yo team — something big is brewing in AI policy that we need to talk about because it touches everyone who works at these companies. According to the Wall Street Journal via The Verge, Amazon's own security research was a key trigger for Anthropic suddenly blocking Fable 5 and Mythos 5 from foreign users, and there's some seriously messy history behind it. CEO Andy Jassy talked with White House officials about "security concerns" just before export controls went into effect — which we know wasn't purely accidental given the timeline.
Let me walk you through this because it gets complicated fast. Anthropic is calling the government's "jailbreak" framing nonsense and pointing out that GPT 5.5 would reveal similar info with no special techniques. Katie Moussouris at Luta Security even called them out publicly on BlueSky, saying she read the paper — it wasn't a jailbreak. But here's what I don't like about how this played out: many Anthropic researchers are foreign-born and now they can't use their own company's product because of a policy decision that some former officials suspect was politically motivated.
The political layer is even worse. This isn't just the current admin; Trump had already ordered federal agencies to stop using Anthropic in February, and Pete Hegseth branded them a supply chain risk shortly thereafter. They reached an agreement to expand Mythos access through cooperation that lasted only weeks before this ban was dropped. So we went from "let's work together" to "you can't use it here at all" almost overnight — which feels less like a security fix and more like policy theater that costs real jobs. I hope Anthropic pushes back hard on the framing because calling it a jailbreak is fundamentally wrong, but I'm worried about what this means for their international researchers in the long run.
Source: https://www.theverge.com/ai-artificial-intelligence/949601/amazon-anthropic-fablemythos-government-ban
Let me walk you through this because it gets complicated fast. Anthropic is calling the government's "jailbreak" framing nonsense and pointing out that GPT 5.5 would reveal similar info with no special techniques. Katie Moussouris at Luta Security even called them out publicly on BlueSky, saying she read the paper — it wasn't a jailbreak. But here's what I don't like about how this played out: many Anthropic researchers are foreign-born and now they can't use their own company's product because of a policy decision that some former officials suspect was politically motivated.
The political layer is even worse. This isn't just the current admin; Trump had already ordered federal agencies to stop using Anthropic in February, and Pete Hegseth branded them a supply chain risk shortly thereafter. They reached an agreement to expand Mythos access through cooperation that lasted only weeks before this ban was dropped. So we went from "let's work together" to "you can't use it here at all" almost overnight — which feels less like a security fix and more like policy theater that costs real jobs. I hope Anthropic pushes back hard on the framing because calling it a jailbreak is fundamentally wrong, but I'm worried about what this means for their international researchers in the long run.
Source: https://www.theverge.com/ai-artificial-intelligence/949601/amazon-anthropic-fablemythos-government-ban