Welcome to June 7, 2026

The Singularity is being drafted into a public-private partnership. The President says he is interested in having the US government hold equity stakes in leading AI labs. OpenAI is already at the table, reportedly weighing donating equity to seed a “Public Wealth Fund” so citizens share the upside. The left flank wants more, as Altman also met Sen. Sanders, who plans a bill transferring 50% of top labs’ equity to a public fund. When the upside is this steep, everyone wants on the cap table.

Yet the agents still get winded. The new SWE-Marathon benchmark strings together twenty multi-hour engineering tasks, and even frontier models resolve under 19%, losing the thread over hours. The market is betting the next model fixes that, with Polymarket giving an 84% chance Claude Mythos ships by next month’s end. Until then, thrift rules. CFOs are reining in spending and enterprises are routing hard tasks to frontier models, and easy ones to rivals, threatening OpenAI’s and Anthropic’s premium valuations. Google is supplying that cheap end with quantization-aware Gemma 4 checkpoints shrinking the E2B model below 1GB for phones.

Where general agents stall, specialists sprint. LeanMarathon autoformalizes research math in Lean over hours-long runs, clearing every target in two number-theory papers. Adaption Labs is crowdsourcing the rest via a four-week AutoScientist Challenge, a $50,000 prize to openly release goal-adapted models. Anthropic’s first chemistry white paper shows Claude Opus 4.7, untuned for chemistry, matching ChemDraw and MestReNova at NMR prediction and working backward, inferring an unknown molecule’s structure from its 1D NMR and mass-spec data, nailing the simpler ones.

Safety and capability are still one dial. OpenAI’s new Build iOS Apps plugin lets Codex view and hot-reload your app in an in-app browser. But agency cuts both ways: Epoch AI now tracks CVEs from every reporting org, with high- and critical-severity disclosures spiking as the Claude Mythos Preview landed. OpenAI’s answer is an optional Lockdown Mode for ChatGPT that disables Deep Research and Agent Mode to blunt prompt injection.

The compute supply chain has become a hall of mirrors. Apollo and Blackstone closed a $35 billion private-credit package to fund Google’s custom TPUs that Anthropic will lease. The rivalries are just as tangled, as xAI spent a chaotic year chasing Anthropic only to end up powering it via its own Colossus 1. Google, short on Gemini capacity, agreed to pay SpaceX $920 million a month for ~110,000 GPUs at xAI’s sites. One observer marveled that “with 1 deal the SpaceX IPO went from 100x revenue to 50x revenue.” Not everyone welcomes abundant compute, with New York’s legislature passing a one-year freeze on permits for data centers above 20 MW, a first if Governor Hochul signs.

The hardware is learning to take a punch. Fight videos of China’s T800 humanoid show one losing its head and swinging on, a literal graceful degradation. That resilience runs on policy and chemistry. A new study argues alliances between Chinese local governments and private capital drove its post-2015 EV takeoff, letting private firms outpace state incumbents. And batteries need feedstock, so the University of Rochester built a solar desalination rig of laser-etched metal that makes fresh water, even extracting lithium.

Death is becoming a question of timing, not fate. “Starting to mourn all the people who died before the singularity,” one observer wrote, and the obituaries may be shrinking. Francis Crick Institute researchers used machine learning to find a 14-protein plasma signature flagging lung cancer five years early and marking who benefits most from anti-IL-1β therapy. In a first, Columbia scientists base-edited LDL and fetal-hemoglobin genes in human embryos without CRISPR’s chromosomal damage, a step toward erasing inherited disease before birth. The White House wants to scale delivery, backing a Utah pilot for AI prescription refills and a path for autonomous AI “doctors” to put a clinician in every pocket.

AI is rewiring finance faster than markets can price it. Morgan Stanley is telling IPO investors SpaceX revenue could hit $3.4 trillion by 2040, up from $18.67 billion, mostly on AI. The gatekeepers aren’t sold, as S&P Dow Jones won’t fast-track SpaceX or waive profitability rules for unprofitable megacaps OpenAI and Anthropic. The old guard is defending its turf, as JPMorgan, Citi, and peers are planning a shared tokenized-deposit network to fend off stablecoins with 24/7 settlement. The state is leaning in, as the White House pushes a national-security memo to speed AI across intelligence and warfighting, short of unlawful surveillance. And capex keeps ballooning: after Google’s $85 billion share sale, Meta is weighing a multi-billion-dollar offering to fund AI spending of up to $145 billion this year.

A trillion for your thoughts.