The US Just Made the Case for Open-Source AI
Monday, 15 June 2026 · 3 min read · Listen to the episode ↗
When the US government pulled Anthropic's consumer AI model Fable from public availability within a day of release after a jailbreak exposed its underlying unrestricted cyber model Mythos, the intervention functioned less as a security victory and more as a global advertisement for open-source AI, with Google Trends recording an all-time high in searches for open-source AI in May.
The US government pulled Anthropic's consumer AI model Fable from public availability within less than a day of its release after a trusted partner of both Anthropic and the US government discovered a jailbreak bypassing the guardrails separating Fable from the unrestricted cyber capabilities of its underlying model, Mythos. Anthropic had itself lobbied to have Mythos regulated as a cyber weapon and marketed it as potentially the most powerful model ever released to the public. When the administration asked Anthropic to fix the jailbreak or take the model down, Anthropic declined, and the White House responded with an export control on Fable. Someone in China was reportedly able to access Mythos through Fable before the intervention. David Saks characterized Anthropic's refusal to fix the security issue as inconsistent with its self-described safety-first position.
A project called Glasswing had already given select institutions, governments, and banks access to a more powerful version of Mythos ahead of full release, and Mythos itself reportedly leaked within the first day or two when users on a Discord server discovered a direct URL to access it. David Canales noted it remains unclear whether the US government response reflected genuine security concerns or whether officials had simply accepted Anthropic's marketing claims about Mythos at face value.
Ben Thompson argued that whether Fable is as powerful as claimed is beside the point, because the next model or the one after will reach that capability level regardless of what happens to this particular release. The more consequential effect of the government's action was that it functioned as an advertisement for open-source AI. Google Trends recorded an all-time high in global search interest for open-source AI in May, with a secondary peak in searches for best open source AI at the end of May, though interest did not hold at those highs following subsequent regulatory actions. The US government is reportedly considering requiring passport upload or know-your-customer verification to access AI models built in the US, restricting access to American citizens, though effective enforcement is complicated by the crypto industry's established ability to route around IP-based geo-blocks. Venice, which uses the VVV token to provide anonymous access to AI models, briefly pumped on news of the Anthropic takedown, but total VVV staked on the platform was approximately 32.5 million as of June 3 and had since declined to under 32 million, meaning the price move did not translate into meaningful platform growth.
The tokenized pre-IPO SpaceX trade across crypto exchanges collapsed after X Stocks, the provider behind tokenized SpaceX offerings with deals at Binance, Bybit, and Bitget, did not receive the volume of SpaceX IPO shares it had expected. Binance's pre-IPO SpaceX campaign attracted 557 million dollars in on-chain subscriptions before being unwound with no allocation distributed to subscribers, and total on-chain customer orders tied to SpaceX access across crypto exchanges exceeded one billion dollars. Kraken did obtain some actual SpaceX shares but was only able to distribute four shares worth of tokenized exposure to subscribers.
Even traditional finance customers at Fidelity and Charles Schwab received fewer shares than requested, as SpaceX reserved a record 20 percent of the deal for individual buyers and retail demand was enormous. Hyperliquid handled the SpaceX IPO by offering only price exposure with no actual share distribution at settlement, and cash-settled perpetuals providing price exposure rather than ownership remain the most functional product currently available to crypto users in this space.
Anthropic has already moved to block tokenized pre-IPO share activity on crypto exchanges, and OpenAI and other frontier labs planning to go public are expected to follow. The on-chain pre-IPO economy and the off-chain post-IPO economy are expected to remain separate until traditional finance and on-chain rails are properly integrated, though over one billion dollars in demonstrated on-chain demand makes Anthropic and OpenAI IPOs likely targets for the next wave of crypto exchange activity.
This summary was generated from the episode transcript and can contain mistakes.