Regulation Archives - The AI Prism

The First Great AI Fund Blowup Just Became an SEC Investigation

The AI Prism Admin — Fri, 28 Aug 2026 10:00:00 +0000

The Securities and Exchange Commission has subpoenaed Goldman Sachs, JPMorgan, Citigroup and Bank of America over their role in the near-collapse of the AI hedge fund Situational Awareness, according to CNBC, citing Reuters. The banks haven’t been accused of anything. But the subpoenas — which seek the fund’s trades, its leverage and its communications with lenders — mark the first time a marquee AI-finance blowup has reached the enforcement stage.

The fund in question went from roughly $45 billion under management to about $10 billion in a matter of weeks in late July (CNBC). Its 25-year-old founder, former OpenAI researcher Leopold Aschenbrenner, was forced to hand his entire public stock portfolio to Ken Griffin’s Citadel at a discount understood to be around 10% (CNBC).

Here’s the part that matters: this wasn’t a story about the AI failing, the models underdelivering, or the technology being a fraud. It was a story about leverage of up to 400% (CNBC), a concentrated bet on a single thesis, and a momentum crash the major indexes barely registered. The SEC isn’t investigating the AI. It’s investigating the plumbing.

Here’s what the fund was, what actually happened in July, why the subpoenas went to banks rather than the fund — and why this moment is about to become the template for how regulators police the AI trade.

A 25-Year-Old Built the Fastest-Growing Fund on Wall Street

Aschenbrenner is German-born and had no prior trading experience when he launched the fund in 2024 (TechCrunch). He became famous for a different reason first: his 2024 essay Situational Awareness: The Decade Ahead, which argued that scaling AI would require a historic build-out of semiconductors, compute, memory and power (the essay). Before the fund, he’d been on OpenAI’s superalignment team, until he was dismissed over what the company described as an improper disclosure of internal information (TechCrunch).

The résumé was the pitch: Columbia valedictorian at 19, enrolled at 15 (TechCrunch). So was the performance. The fund returned 439% through June of this year, per the Financial Times (TechCrunch), and peaked at an estimated $45 billion in assets (CNBC).

The early backers read like a who’s who of AI-adjacent capital: quant giant Jane Street, Stripe co-founders Patrick and John Collison, and former Meta executives Daniel Gross and Nat Friedman (TechCrunch). They weren’t betting on a traditional hedge fund. They were betting on the thesis — that AI infrastructure was the trade of the decade.

The Trade: Chips Long, Software Short, Four Times Leverage

Situational Awareness ran one of Wall Street’s most crowded trades with a twist. On the long side sat concentrated positions in the companies expected to supply the AI build-out: memory maker SK Hynix, neocloud CoreWeave, storage maker SanDisk, plus Nebius, Bloom Energy, SharonAI and IREN, per public filings as of March 31 (CNBC).

On the short side: software names like Adobe, on the theory that AI would eat their moats (CNBC). The structure looked like a natural hedge — AI winners long, AI losers short. Then came the leverage.

CNBC reported the fund ran on leverage of up to 400% (CNBC), and the Wall Street Journal reported an options overlay — a so-called “Texas hedge” of long stock against short calls — that capped upside while amplifying downside (WSJ).

It was also the consensus trade. Over 80% of fund managers in Bank of America’s monthly survey named “long global semiconductors” the most crowded trade in the market (The Economist). When everyone’s in the same boat, the boat doesn’t have to sink — it just has to rock.

A Momentum Crash the Indexes Never Showed

Here’s the counterintuitive part: the S&P 500 was near record levels while Situational Awareness was being destroyed (CNBC). The damage was hidden inside one of the fastest reversals in market history — “the largest/fastest momentum crash in modern history,” per BTIG’s chief market technician Jonathan Krinsky (CNBC). Morgan Stanley’s sector-neutral Momentum Index tumbled 17.4% in four trading days — its worst such decline on record, worse than the dot-com bust, the pandemic shock and 2022 (CNBC).

The fund’s longs fell 50% to 78% from their peaks by July 29, while its short leg — software — rallied (CNBC). Both sides lost at once. The hedge didn’t hedge.

The AI trade had been faltering since June — even with major indexes flat, two of the fund’s biggest longs, SanDisk and Bloom Energy, tumbled more than 50% (CNBC). Margin calls followed, then forced selling, then a deleveraging spiral: a shrinking equity cushion, more collateral demanded, more positions dumped into a falling tape (CNBC). The fund’s July 24 letter called the selloff the best buying opportunity since early last year and invited fresh capital from August 1 — the appeal drew less than hoped, per Bloomberg (TechCrunch).

It was a liquidity crisis, not a returns crisis: down 67% in July, the fund was still up about 80% on the year (WSJ). In its investor letter, the fund blamed short sellers and compared the episode to a bank run (WSJ via HN).

Citadel Bought the Whole Book at a Discount

By July 30 the forced seller had a buyer. Ken Griffin’s Citadel reached a deal to buy the fund’s public portfolio at a discount understood to be around 10% — after entering discussions on July 29 (CNBC). It’s a familiar pattern for the firm: stepping in to buy quality assets from leveraged players forced to unwind (TechCrunch).

The trade worked again. By August 21, Griffin’s letter to clients said Citadel had unwound more than 80% of the acquired risk through more than 100 block trades worth over $4 billion (CNBC). Its flagship Wellington fund finished July up 5.94% — the best month since 2022 (CNBC).

Griffin credited the banks for the speed of the handover: “A transaction of this magnitude could not have been completed without the extraordinary cooperation of the trading and prime brokerage teams at the banks serving both firms” (CNBC). Sit with that quote for a second — the same banks now fielding SEC subpoenas were the ones midwifing the fire sale.

The sale also marked the bottom. AI infrastructure stocks rebounded the day Citadel stepped in, and SK Hynix and CoreWeave have rallied since (CNBC, CNBC). Shares of several AI infrastructure companies jumped the moment the deal surfaced, instantly boosting the value of the positions Citadel had just acquired (CNBC). Bank of America CEO Brian Moynihan — whose bank was one of the fund’s prime brokers — said BofA would have been “fine” even without the rescue (CNBC).

The Contagion That Almost Was

One fund blowing up is an anecdote. What nearly happened around it is why regulators care. Jane Street — a Situational Awareness investor and one of the biggest names in market making — took a $15 billion hit in July, including losses tied directly to the meltdown, per FT and Reuters (FT, Reuters). It still generated more than $40 billion in net trading revenue over the past year (FT).

The Economist spelled out the tail risk: if Citadel hadn’t stepped in, the fund might have had to fire-sell tens of billions of dollars of assets — assets other firms held leveraged positions in. That could have triggered a cascade of margin calls, and potentially a credit or even bank crisis (The Economist). Columnists invoked Long Term Capital Management, the 1998 hedge fund collapse that nearly took the financial system with it (Economist via HN).

Michael Burry of The Big Short fame was less worried about the system and more about the trade: on the rebound day he added bearish positions in Micron, the VanEck Semiconductor ETF and Nvidia puts, calling the reversal “historic… even more so than what happened 26 years ago” (CNBC).

The next stress point, according to analyst Eric Newcomer, may be Nvidia’s vendor financing — arrangements that help customers pay for the chip giant’s own GPUs (Newcomer). When the collateral is the product, a margin call becomes a demand for hardware.

Why the SEC Went After the Banks First

The New York Times first reported the subpoenas on August 24 (NYT). Per Reuters, the SEC is seeking information on the fund’s trades, use of leverage and communications with the investment banks — Goldman, JPMorgan, Citigroup and Bank of America, the lenders that supervised its trading and channeled funding to it (CNBC, TechCrunch). The agency reportedly told the banks to “preserve any information” about the fund (TechCrunch).

Read the scope carefully. The SEC is not, publicly, investigating the AI thesis. It’s investigating the financing — who extended how much leverage, on what terms, with what visibility into a book that was levered four times into a crowded trade. The fund itself was already in the agency’s orbit: its holdings are on file via quarterly 13F reports (last10k). What the SEC wants now is the part that never made it onto the filings. The breadth of the request — trades, leverage, communications — suggests the agency is reconstructing how the fund was financed stage by stage, and what each lender knew about the leverage at every step.

None of this means wrongdoing. CNBC notes such inquiries routinely conclude without enforcement action (CNBC). The fund’s statement struck a cooperative tone: “It is to be expected that regulators would closely examine any funds that are high profile, produce significant returns, or have particularly dramatic drawdowns… We are a highly-regulated business and will cooperate to the fullest extent with any regulatory request” (CNBC).

The real question is what the SEC learns about the banks’ conduct — whether prime brokers competed so hard for the fund’s business that the usual guardrails loosened. Moynihan hinted at the reflex on August 5: “The tendency is to tighten the underwriting standards, just a hair” (CNBC).

What the First AI-Finance Blowup Actually Proves

First, it doesn’t prove the AI trade was a bubble. The same stocks the fund was forced to dump have since rallied — the crash was a positioning event, not a fundamentals event (CNBC). Whether the broader AI economy is a bubble is a separate argument, one we’ve made at length in our analysis of what survives when the AI bubble bursts.

What the blowup does prove: leverage has quietly become the AI trade’s connective tissue. The fund was the most visible example of borrowed money amplifying an AI thesis, and its near-collapse “cast light on the various ways in which leverage is increasingly underpinning the wider AI boom” (CNBC).

Moynihan was blunter the week after the collapse, calling the episode one of “these… warning shots”: “Valuations get out, leverage in the system gets there. You have to be careful” (CNBC). His comments suggested the largest prime brokers were already reexamining their exposure to highly leveraged investment firms (CNBC).

Second, the players are already moving on. The fund kept its private assets, including an Anthropic stake valued around $5 billion — Anthropic was last valued at $965 billion in May, with an IPO expected as soon as October (TechCrunch). Other private holdings include chipmaker MatX and AI data-center startup Fluidstack (TechCrunch). In early August the fund put $400 million into chip startup Source Foundry, bringing its total there to $500 million (TechCrunch). Bloomberg reports investors still clamor to back the “AI whiz kid” (Bloomberg).

Third, this is now the template. The SEC’s bank-first approach — subpoena the lenders, map the leverage, see who looked the other way — will likely be the playbook for the next AI-finance stress event. Given how much of the AI economy now runs on financed capacity, there will be a next one.

The Bottom Line

Situational Awareness wasn’t the first leveraged fund to blow up, and it won’t be the last. What made it different was the costume: the AI thesis was so compelling that hundreds of millions of dollars and four times leverage felt rational. The subpoenas are a reminder that when the trade is crowded and the money is borrowed, the thesis is never the whole story — the plumbing is.

The hype-fund that nearly collapsed is now an SEC investigation, and the first big AI-finance blowup wasn’t the AI’s fault — so what happens to the next fund that bets everything on AGI?

References

The post The First Great AI Fund Blowup Just Became an SEC Investigation appeared first on The AI Prism.

The AI Lobbying Explosion: Record Spending Is Reshaping Washington

The AI Prism Admin — Tue, 04 Aug 2026 20:05:04 +0000

The Price of a Seat at the Table

Here’s a number to sit with: in the first half of 2026, Anthropic nearly tripled its federal lobbying spending to $3.53 million. OpenAI roughly doubled its own to $2.22 million — a record for the company. Those are the figures from federal disclosure filings, reported by the Financial Times and picked up across Hacker News with 277 points and 144 comments.

Two companies that didn’t exist a decade ago are now writing checks to influence the people who write the rules for the most consequential technology since the internet. That’s not news in itself — every industry lobbies. What’s new is the slope of the curve.

AI lobbying didn’t grow incrementally. It exploded. In 2023 alone, according to OpenSecrets data compiled by CNBC, lobbying by AI companies jumped 185% — from 158 organizations to more than 450. Combined federal spending by those organizations crossed $957 million. Nvidia, OpenAI, Anthropic, Palantir, ByteDance and Tesla all registered as lobbyists for the first time that year.

The AI industry has discovered that the fastest way to shape its future is no longer a better model — it’s a better-connected law firm.

From Zero to Seven Figures in Three Years

OpenAI’s own trajectory is the cleanest case study. In 2023, the company spent $260,000 on federal lobbying. In 2024, that figure jumped to $1.76 million — nearly seven times more, per MIT Technology Review. In the first half of 2026, it hit $2.22 million. If the second half matches, the company will have grown its lobbying budget roughly 17x in three years.

What changed between 2023 and 2024? The answer is visible in the résumés OpenAI started collecting.

Chan Park, former counsel to the Senate Judiciary Committee and a Microsoft lobbyist. Reginald Babin, former counsel to Senate Majority Leader Chuck Schumer. Meghan Dorn, former staffer for Senator Lindsey Graham. Matt Rimkunas, a veteran of the energy investment world. Chris Lehane, the political operative who ran Al Gore’s 2000 campaign and later Airbnb’s policy machine.

That’s not a government affairs team. That’s a shadow cabinet.

The hires tell you exactly where the industry thinks its future is decided: not in the lab, not in the marketplace, but in the corridors where energy policy, national security and defense budgets get written. OpenAI’s pivot from safety messaging toward energy, national security and defense — including its reported partnership with defense contractor Anduril — is the policy strategy made flesh.

The Revolving Door Is a Two-Way Street

Washington’s revolving door has always spun, but the AI era has made it spin at model-training speed.

The pattern is consistent across the frontier labs: hire people who just wrote the laws, or who work for the people who write them. Anthropic’s 2026 expansion reportedly included Ballard Partners, the lobbying firm founded by a former Trump campaign finance chair and now connected to the administration — a sign, per Bloomberg’s reporting, that the company is building relationships on both sides of the aisle and both sides of the transition.

The hires cut both ways. Every former Hill staffer who joins an AI company brings two assets: relationships and knowledge of where the bodies are buried in pending legislation. That’s precisely why the industry is willing to pay top dollar for them.

The result is an information asymmetry that has nothing to do with AI capability. When an AI company’s lobbyist used to draft the AI bill, the company doesn’t need to read the bill to know what’s in it — they already know who wrote which sentence.

What the Money Actually Buys

Lobbying isn’t corruption; it’s access. But the returns on that access are visible in the legislative record.

Take Europe. In 2023, documents obtained by TIME through FOIA requests showed OpenAI lobbying the EU to water down the AI Act, arguing that its GPT-3 model shouldn’t be classified as “high risk.” The argument’s fingerprints are visible in the final text of the regulation — the EU’s flagship AI law ended up with carve-outs and a phased approach that the industry pushed for.

The 2026 calendar is full of similar stories:

May 2026: Tech-industry lobbying helped block a Trump administration executive order on AI, per the Washington Post — the rare case of an industry killing a rule it didn’t want, rather than shaping one it did.
March 2026: The EU’s “Digital Omnibus” package reflected big-tech messaging almost point for point, according to observers of the Brussels process.
July 2026: Uber — now an AI company in its own right — lobbied New Jersey on a rule that would require 85% of robotaxi miles to have a human safety driver, a threshold its competitors couldn’t meet and a textbook example of using regulation as a moat.
December 2025: An Arizona city rejected a proposed data center after an AI-industry lobbying push for tax breaks backfired in public, per Politico — the rare case where the playbook failed.

None of these are scandals. All of them are the system working exactly as designed. The question is whether that design serves the public interest when the technology being regulated is moving faster than the legislative branch can type.

The Safety Movement Shows Up Late — and Poorly Funded

Here’s the asymmetry that should worry everyone who thinks AI needs guardrails.

The frontier labs spend millions on lobbying, and we have covered the AI safety debate driving that spending. The organizations arguing for safety and regulation? In late 2023, the Center for AI Safety and the Center for AI Policy registered their first lobbyists with roughly $100,000 in spending each, per Politico — funded largely by Open Philanthropy and Lightspeed Grants.

Do the math. OpenAI spent $1.76 million lobbying in 2024 — 17 times what both major safety organizations combined spent in their first year. The safety movement isn’t losing the policy war because its arguments are weak. It’s losing because it’s showing up to a spending war with a slingshot.

The frontier labs don’t need to win every argument. They just need to make sure the arguments that matter happen in rooms where they have a seat.

The Defense Pivot

Watch what happens when an AI company’s lobbying shifts from one theme to another — that’s the roadmap for where the money is heading next.

OpenAI’s disclosure history shows exactly this pivot. In 2023, the company’s public posture and lobbying centered on safety, responsibility, and the benign framing that helped it land EU exemptions. By 2024 and into 2025, the emphasis had moved to energy, infrastructure, national security and defense, per MIT Technology Review’s analysis. The Anduril partnership and the company’s positioning around military applications weren’t product decisions alone — they were policy plays that aligned the company with the two budgets that never shrink in Washington: defense and energy.

This is the mature playbook. When a technology becomes strategically important, its companies stop lobbying for permission and start lobbying for contracts. The AI industry has reached that stage years earlier than most sectors because its infrastructure needs — data centers, grid capacity, chips — are themselves national-security questions.

Who’s Not in the Room

It’s worth listing who the record spending does not represent.

Civil society organizations working on AI accountability have almost no lobbying presence. Academic researchers who study AI risks publish papers, not disclosure filings. Labor groups representing the workers AI is expected to transform have only begun to organize around the issue. And the safety organizations that did register lobbyists — the Center for AI Safety and the Center for AI Policy — started with roughly $100,000 each, a rounding error next to a single quarter of Anthropic’s spending.

The asymmetry has a structural cause: lobbying is an investment, and the people most affected by AI policy have no financial return to capture. A company that spends $2 million to shape an AI law can expect that law to protect billions in market value. A worker whose job is transformed by that same law gets no equivalent payoff for opposing it. So the spending concentrates where the returns concentrate, and the conversation narrows accordingly.

The Stack Beneath the Headlines

The AI-specific numbers are dramatic, but they’re a rounding error compared to the broader tech lobbying machine they’re joining.

Look at the 2025 disclosure data: Meta spent a record $26.29 million on federal lobbying. Amazon spent $18.9 million. Alphabet $16.5 million. The U.S. Chamber of Commerce — which fights AI regulation on behalf of its members — spent $72.1 million. The tech sector as a whole runs around $450 million a year in federal lobbying, third overall behind only the biggest industrial sectors.

The AI companies aren’t inventing a new playbook. They’re buying into an existing one, at scale, with the urgency of a technology that knows its regulatory window is closing. Every dollar spent today is an investment in which version of the AI rules gets written — and which version gets buried.

What This Means for the Rest of Us

There are three consequences worth naming, none of them conspiratorial.

First, regulation will lag capability — permanently. Not because regulators are lazy, but because every legislative proposal now goes through a gauntlet of well-funded expert pushback that didn’t exist two years ago. By the time a rule passes, the technology has moved two generations past what it regulates. The EU’s AI Act took four years to negotiate; the models it was written for are already obsolete.

Second, the public conversation is being outsourced. When the people writing the first drafts of AI laws are former staffers of the people funding them, the range of “reasonable” policy options narrows. Options that threaten the business model get filtered out long before they reach a vote. State-level AI bills are where this shows up first — dozens of them get introduced each session, and the ones with the most lobbying attention are the ones that get quietly rewritten or shelved.

Third, the gap between corporate AI power and public understanding is widening. The average person experiences AI as a chatbot. The industry experiences it as a policy war. Those two realities are drifting apart, and the drift is being financed at $2 million a quarter.

What to Do About It

This isn’t a call to despair — it’s a call to pay attention. A few things worth doing:

Follow the disclosures. Lobbying data is public. OpenSecrets and the Senate’s LDA database are free. Knowing who spends what is the first step to knowing whose voice is loudest.
Fund the other side. The safety organizations that registered lobbyists in 2023 are outspent by an order of magnitude. If you believe in oversight, the most effective donation you can make is to the people arguing for it in rooms with the people writing laws.
Ask your representatives about AI — specifically. Generic questions get generic answers. Ask which AI bills they’ve read, who they’ve met with, and what their position is on training-data disclosure. The answers tell you whose office is listening to whom.
Read the fine print of “AI for good.” Every corporate announcement about responsible AI should be read alongside the lobbying disclosure. The two together tell the real story.

The Bottom Line

AI companies are spending record sums on lobbying because it works. The returns are visible in every watered-down rule, every blocked executive order, every carve-out that made it into law.

This isn’t a morality play. It’s the normal operation of a system where the people with the most at stake get the most say. The problem is that with AI, the stakes aren’t just corporate — they’re civilizational, and the rest of us are showing up to that fight unrepresented.

When a seat at the table costs $2 million a quarter, the real question isn’t who’s at the table. It’s who isn’t.

So here’s the question for your representatives: When the last AI bill was drafted, whose lobbyists were in the room — and whose weren’t?

References

The post The AI Lobbying Explosion: Record Spending Is Reshaping Washington appeared first on The AI Prism.