TWO WEEKS THAT SHOOK THE WORLD: On 12 June, the US government shut down access Anthropic's Fable 5 frontier AI model. The AI ecosystem suddenly understood 'sovereign risk'. That was the shot. Here comes the chaser... On 26 June, OpenAI CEO Sam Altman let it be known that the US government had asked the firm to strictly control the rollout of their forthcoming GPT-5.6 model. Altman told staff the government would be "approving access customer by customer during this preview period" for GPT 5.6, according to The Information. Read between those lines: if the government likes you, access will be granted to the latest-and-greatest AI tools. And if not, well...what could possibly go wrong? 🤔 Kate Carruthers John Allsopp Zina Kaye Snow x Iain Chalmers Rachel Chalmers Zoë Vaughan Zac Pullen Viveka Weiley 🦉 Joey Meynink Anthony Saba Simon Kriss Murray Hurps David Chambers Drew Smith Sally Dominguez Catherine Ball Andrew Charlton Maria MacNamara Hugo O'Connor Chris K. David Frayne Chris Rickard https://lnkd.in/eRFTcNjq
Wow it is almost like that state control they have always kvetched about 👀
I think this in effect represents a 6 to 18 month reset on progress as that is how far behind the best Open Weight models seem to be to me. That said where we are is good and workable. Perhaps an approach is harness and pipelines that combine disparate Open Weights models in testably effective ways - a kind of model pair programming. If the models we will not get access to are 2-4x more infra expensive anyway, maybe we can use 2 differing models in parallel, one looking over the others shoulder and giving feedback rapidly, maybe at the deeper level of monitoring logits like is used for speculative decoding ?
Simon Kriss
2w
Madness that most cannot see the burning bridge in front of them (even as it creaks beneath their feet).
Sam Sabey
2w
As we are separated into the haves and have nots while the option that remains carries its own questions. Either way I feel like we may be approaching the end of the rush of 2026 and towards the shift towards achieving similar outcomes from new and varied inference. The "Hormuz" moment in AI
The gap between policy and practice can only be narrowed if those setting the rules care more about the world as it will be and skill up to challenge what’s proposed. People like Julie Inman Grant PSM set the example.
it's fucking bananas. ... but i'm also not surprised 😞
Export controls have had these restrictions since 2024 in Australia. However there is a wider issue about restricting individual empowerment, which increasingly seems to be part of the sub-context of legislation around the world over the last 5 years.