Amazon Researchers Report Fable Jailbreak To Anthropic Before Alerting US Officials

Source: https://www.ft.com/content/f6940d59-28f4-4ae4-a569-c6fc421e52b9

@ns123abc And yet no one has claimed the open $25,000 bounty that OpenAI created for a universal jailbreak of GPT 5.5

@ns123abc This is the best advertisement OpenAI can get. Honestly, Fable 5 is on a completely different level than ChatGPT 5.5. I’m totally for Mythos and Fable 5 being restricted to American citizens for a set time.

@ns123abc OpenAI reading this:

Wait, why am I in it?

@ns123abc are all technology revolutions this dramatic

@ns123abc They 'cried wolf' a lot about others then got hit hard.

@ns123abc derp

@ns123abc This reads to me as: Amazon saw how capable Fable is, liked having exclusive access for its business, didn't like all the normies having the same access they had, and so raised a flag that couldn't be ignored.

@ns123abc Amazon invested $13bn in Anthropic. Then Andy Jassy took their jailbreak report to the US government. Then the government banned Anthropic’s best model. That’s a fascinating use of a $13bn investment.

@ns123abc "The capability exists in OpenAI's models too" is probably in the original Amazon security report, the part that didn't make it into the government briefing is the interesting part

@ns123abc Responsible disclosure went: researcher → Anthropic → CEO → government officials → export ban, At some point in that chain it stopped being about the vulnerability