Overview
This column’s first installment covered the strange disappearance of Claude Fable 5, the most capable AI model on the market, forced offline by a U.S. export-control order days after launch. Less than three weeks later it un-vanished. The Commerce Department lifted the order on June 30, and Fable came back worldwide on July 1. The return has conditions. Anthropic agreed to hunt for its own security flaws, coordinate future launches with the government, and report malicious use. Its new safety filter was tested by Commerce’s own AI lab.
The order rested on a jailbreak that got the model to flag security flaws in code and, in one case, show how a flaw could be abused. Dean Ball, an AI policy expert who served for a time in the Trump administration, could not tell whether the move was “lawfare against Anthropic in particular or extreme national-security hawkery.” He called it “simply cartoonish.”
Fable’s return is not a return to normal. Around the same time, OpenAI launched GPT-5.6 Sol, the flagship of its new three-tier family (Luna, Terra, and Sol), to about 20 White House-approved organizations while a formal review process takes shape. The list is not public, and there is no application to join it. OpenAI said it does not believe “this kind of government access process should become the long-term default.” Zvi Mowshowitz, whose AI newsletter is required reading on these episodes, was blunter. “Ad hoc opaque politicized decisions from the White House on who gets frontier intelligence, however, is Not The Way.”
You can feel the safety machinery in ordinary work. I’ve had Fable 5 balk at bland patent questions because the patents covered pharmaceutical products. After its return, Fable seems even touchier than before.
For lawyers weighing the new lineup, two comparisons matter. On capability, Mowshowitz rates Sol “a substantial improvement over GPT-5.5, but still short of Mythos,” Fable’s restricted sibling. On price, Sol runs $5 in and $30 out per million tokens, about half of Fable’s $10 and $50.
That price gap tells you what to expect in the legal tools. Harvey runs a mix of models, and its newest addition is Claude Sonnet 5, not Opus, and certainly not Fable. CoCounsel’s rebuild, covered here last week, says it runs on Claude without naming the model. It is vanishingly unlikely to be Fable. Given the prices, do not expect Fable inside the products most firms buy anytime soon.
Fable is more a sign of what is coming than a tool most lawyers will use directly today. But lawyers who learn what the frontier can do now will recognize it when it reaches cheaper tools in the future.
The Weekly Inference is a recurring Step Into IP feature. Nothing here is legal advice.
