June 2026 newsletter

The newsletter frames export restrictions as a persistent headwind, but our archive shows those restrictions on Fable 5 were already lifted by July 1, just two days before publication. Willison's framing may be slightly behind the regulatory curve, which matters because the bifurcation thesis he builds partly rests on a policy environment that had already shifted.

Multiple stories from July 1 fill in the gap directly. The Decoder's coverage of Fable 5's return confirmed the two-week suspension ended after Anthropic deployed a new safety classifier following a jailbreak discovered by Amazon researchers. Ars Technica framed this as a precedent: structured safety evaluation can satisfy government concerns rather than trigger indefinite holds. That context reframes Willison's regulatory headwind argument. The restrictions weren't a structural barrier so much as a negotiated checkpoint, and Anthropic's compliance posture appears to have been the deciding variable.

If GLM-5.2 posts competitive scores on the same evals used to benchmark GPT-5.6's three-variant lineup (per The Decoder's July 1 paper coverage), that would confirm the open-weights convergence Willison flags. If GLM-5.2 underperforms on those specific splits, the China capability acceleration story needs more scrutiny.