The Once And Future Fable #5
Zvi's AI Roundups · Zvi Mowshowitz · 2026-06-30
Zvi Mowshowitz surveys the week's major AI developments including the partial restoration of Claude Mythos 5 access to US institutions, SCOTUS's ruling eliminating independent AI regulatory bodies, Alibaba's alleged distillation attacks on Claude using 25,000 fraudulent accounts, and the open-weight model safety debate.
Appears in
Extraction
Topics: ai-policyus-ai-regulationopen-source-aiai-safetyanthropic-claudeexport-controls
Claims
- Claude Mythos 5 has been restored to more than 100 US institutions but Claude Fable 5 remains restricted pending NSA and Pentagon approval.
- SCOTUS's 6-3 ruling in Slaughter v. Trump, overturning Humphrey's Executor, makes creating an independent federal AI regulatory body legally impossible under current doctrine.
- Anthropic formally accused Alibaba of distillation attacks on Claude using approximately 25,000 fraudulent accounts to generate training data.
- CAISI has been under a stop-work order throughout the Mythos and Fable access restriction period, leaving the White House without in-house AI technical expertise at the critical moment.
- Open-weight model advocates explicitly want models free from monitoring and safety guardrails, and proponents dismissing the resulting risks are not engaging seriously with AI safety.
Key quotes
De facto ad hoc improvised involuntary preapproval of model releases.
Anthropic has formally accused Alibaba of massive distillation attacks on Claude, via almost 25,000 fraudulent accounts generating outputs for this purpose.
I assure you that no, you do not want them to dump Mythos onto Hugging Face, you would not like what happened to your life next.
What they're big mad about is the idea that this could also have downsides. If such AIs are Sufficiently Advanced AIs, then that implies the ability to do a number of potentially highly dangerous things.