A Big Shift In the AI Race

Jun 17, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
June 17, 2026

HEADLINES

The Anthropic Crisis, Day Five

Five days in, the US government's shutdown of Mythos and Fable shows no signs of resolution. On Monday, Anthropic sent a largely technical delegation to Washington — Chief Compute Officer Tom Brown, Head of External Affairs Sarah Heck, Head of Red Teaming Logan Graham, and Senior Security Researcher Nicholas Carlini — but none of the key decision-makers who appear to have actually driven the ban (National Cyber Director Sean Cairncross, Treasury Secretary Bessent, or Chief of Staff Susie Wiles) were in the room. Commerce Secretary Lutnick phoned in from the G7 in France. Absent the people with real authority, it's questionable whether any agreement could have been reached even if the talks had gone well.

The WSJ profiled Carlini as a respected skeptic of AI cybersecurity claims who changed his mind after getting access to Mythos — he found critical bugs in both Ghost and Linux that he'd never previously been able to crack, and told a standing-room crowd of experts that the attacker-defender balance of the past two decades "seems like it's probably coming to an end." That context matters: Carlini was reportedly among those inside Anthropic pushing not to release Mythos back in March.

It's worth underscoring that this is not just a Fable ban — Mythos is the bigger issue. Sources say the Commerce Department was willing to let Fable come back online if Anthropic fixed the jailbreak, but Mythos is a separate question entirely. The Washington Post reported that Anthropic added around 50 firms to Project Glasswing without providing the government the list for days, and once officials had it, they found a South Korean telecom company suspected of ties to the Chinese government. The jailbreak appears to have been the trigger; the expanded Glasswing access, the root cause.

On Tuesday, Bloomberg published the full text of Commerce's letter — essentially a threat of criminal and civil penalties requiring Anthropic to remove access for all foreign nationals. During his Friday night call, Dario Amodei reportedly told Lutnick "This means we can't have the model out." Lutnick's reply: "That's the point." Legal analyst Charlie Bullock of the Institute for Law & AI called the legal theory "strange, very aggressive, and probably vulnerable to challenge," while arguing that the real problem is the lack of any actual legislative framework — bad for Anthropic, bad for the administration, bad for everyone. The cybersecurity community, geopolitical analysts, and the business world are all sounding alarms. Helen Toner, former OpenAI board member, put the technical reality plainly: "It is a pretty widely agreed upon fact that you cannot fully fix jailbreaks in these models." As of this morning, Dario Amodei and other AI CEOs have traveled to France for the G7, where European leaders — freshly locked out of Mythos after just weeks of access — will reportedly meet with Trump and Amodei in an attempt to find a resolution.

MAIN STORY

A Big Shift in the AI Race

Even as Anthropic dominates the headlines, a separate realignment is underway in the AI race itself. The economics of mature agentic AI are starting to clarify — companies are grappling with what full agentic workloads will actually cost, scrambling to become more token-efficient, and watching the broader macro implications of the AI infrastructure build-out. Into this liminal moment steps Elon Musk, who is using SpaceX's newly minted IPO war chest and sky-high valuation to make a series of moves that could dramatically reshape the competitive landscape.

CURSOR PUTS SPACEX BACK IN PLAY

SpaceX Post-IPO
The stock keeps climbing — and the implications keep compounding.
SpaceX closed Tuesday at $201.80, up 49% from its IPO price, putting the company at a $2.6 trillion valuation and briefly ahead of Amazon. Most of Elon Musk's wealth remains locked up in his 46% stake, so the trillionaire headlines are somewhat notional — but the strategic implications of a surging stock price are entirely real. As Flexport CEO Ryan Petersen pointed out, Elon made more in a single day of SpaceX gains than Warren Buffett made in his entire career. Some argue SpaceX left money on the table. Others flag that with only 5% of the stock currently tradable and over 50% unlocking by Christmas, bearish pressure could build. For now, though, the strong debut is seen as a green light for OpenAI and Anthropic's own IPO ambitions — with Wedbush analyst Dan Ives calling it "an important watershed moment for the broader tech sector."

SpaceX Acquires Cursor
A $60B all-stock deal that changes the shape of the AI race.
SpaceX has finalized its acquisition of Cursor for $60 billion in stock — and the strategic significance extends well beyond the price tag. Cursor hit $4B in ARR growing 7x year-over-year, but the real story is what it means for SpaceXAI's model ambitions. At the Compile event this week, Cursor CEO Michael Truell teased a new 1.5 trillion parameter model trained from scratch on over 100,000 GPUs at SpaceX's data center campus — the first time Cursor has had access to genuine frontier-scale compute. Previous Composer models were post-trained on a Kimi base at a fraction of that scale; this is a different category of bet. Truell said the 10–20x compute scale-up "really lets us get to frontier and hopefully soon leapfrog it." On distribution, the calculus is simple: SpaceXAI has probably added more enterprise users in a single day than in two and a half years of Grok. Meanwhile, the WSJ argues this may only be the beginning — SpaceX's $86B IPO war chest wasn't touched for this all-stock deal, giving Elon substantial resources for further acquisitions. Bill Ackman summed up the dynamic: "Value begets value. Talent begets talent." The FT drew the comparison to Facebook's Instagram acquisition: instead of competing with the hot new thing, just take it off the board early.

DOJ Intervenes in xAI Data Center Lawsuit
The US government calls Grok vital to national security — and confirms AI's role in Iran targeting.
The DOJ joined xAI's motion to dismiss an NAACP lawsuit arguing that Colossus II's unpermitted gas turbines violate the Clean Air Act. The government's filing stated that shutting off power to xAI would threaten "national, economic, and energy security." A supporting Pentagon filing confirmed that Grok was used for targeting decisions in recent Iran strikes — enabling, per the official, "over 2,000 munitions to 2,000 distinct targets within 96 hours." Grok isn't alone: Anthropic, Google, and OpenAI models are also cleared for classified use, and Dario Amodei recently confirmed Claude was used in missile targeting, though he said he didn't know the specifics. The juxtaposition isn't lost on observers: the same week the government pulled Anthropic's models for foreign nationals, the DOJ went to court to protect xAI's data center as a national security asset.

OpenAI Financials Leak
Ed Zitron cries doom; the FT tells a more complicated story.
AI skeptic Ed Zitron published OpenAI's fully audited numbers ahead of the company's IPO. The headline figures: $3.7B revenue and a $5B net loss in 2024; $13B revenue and a $38.5B net loss in 2025. Zitron called the losses "astronomical" and the financial condition "deeply concerning." The FT, to whom Zitron also shared the numbers, arrived at a very different conclusion — noting that $30B of those 2025 losses were a one-time non-cash accounting charge tied to OpenAI's conversion to a public benefit company. Strip that out along with stock comp and OpenAI lost $8B in 2025, which is not nothing but is a different story. The figure that got less attention in either framing: OpenAI is generating solid gross margins on inference — $3.7B in revenue against $2.7B in direct costs in 2024, and $13B against $7.5B in 2025. The Information reports OpenAI burned $3.7B in Q1 2026 (excluding $8.6B in training costs), but has $73B in cash on its balance sheet, up from $40B in December. Sam Altman has been careful not to commit to an IPO timeline, and it's easy to construct a scenario — SpaceX unlocks, market volatility, continued Washington turbulence — where staying private a bit longer makes sense.