The U.S. Securities and Exchange Commission has subpoenaed several major Wall Street banks as it examines their dealings with Situational Awareness, the AI-focused hedge fund that suffered a dramatic collapse in assets last month after a technology-stock sell-off triggered a series of margin calls.

The subpoenas seek information about the fund’s trading activity, leverage and communications with its investment banks, including Goldman Sachs, JPMorgan, Citigroup and Bank of America, according to Reuters, citing a person familiar with the matter.

The regulatory inquiry follows one of the most severe recent episodes of forced deleveraging tied to the AI investment boom. Situational Awareness, led by former OpenAI researcher Leopold Aschenbrenner, saw its assets plunge from about $45 billion to roughly $10 billion in late July after losses on concentrated technology positions triggered margin calls from several prime brokers.

The fund was subsequently forced to unwind a large portion of its publicly traded portfolio, including positions in semiconductor maker SK Hynix and AI data-center company CoreWeave.

The episode has brought increased attention to the role leverage has played in the rapid expansion of AI-related investments, especially as hedge funds and other institutional investors have built large positions in companies benefiting from surging demand for AI infrastructure.

Situational Awareness reportedly used leverage of as much as 400%, meaning relatively small changes in the value of its underlying positions could produce disproportionately large gains or losses.

When technology shares declined sharply in July, that leverage became a major vulnerability.

The resulting margin calls forced the fund to sell assets into a falling market, accelerating the unwinding of its positions and contributing to the sharp contraction in its portfolio.

Citadel, the multistrategy hedge fund founded by Ken Griffin, stepped in to purchase the positions at a discount understood to be around 10%. Griffin said in a letter to investors Friday that Citadel had since reduced about 80% of the risk associated with the portfolio.

The shares that were at the center of the forced liquidation have subsequently recovered, with both SK Hynix and CoreWeave rallying.

The development exposes the difficulty of unwinding large leveraged positions when multiple financial institutions are simultaneously exposed to the same trades. Prime brokers provide hedge funds with financing, securities lending, trading infrastructure and other services that allow them to take positions substantially larger than the amount of capital they have invested. Those relationships can amplify returns during periods of rising markets but can also force rapid asset sales when collateral values fall.

The SEC’s subpoenas appear aimed at understanding how those relationships operated in Situational Awareness’ case, including how much leverage the fund used, the scale of its positions and how its prime brokers communicated with the firm as losses mounted.

The inquiry does not mean that the banks or Situational Awareness have been accused of wrongdoing.

Regulatory information requests can ultimately conclude without enforcement action, and the SEC’s investigation may simply be an effort to understand how the losses developed and whether risks were adequately managed.

“It is to be expected that regulators would closely examine any funds that are high profile, produce significant returns, or have particularly dramatic drawdowns,” Situational Awareness said in a statement.

“We are a highly-regulated business and will cooperate to the fullest extent with any regulatory request.”

The inquiry nevertheless puts a spotlight on a broader issue confronting financial markets: how much leverage has accumulated around the AI trade.

AI has become one of the dominant investment themes in global markets, attracting enormous amounts of capital into semiconductor manufacturers, cloud companies, data-center operators and other infrastructure providers. The scale of the opportunity has encouraged investors to take increasingly large positions in companies expected to benefit from AI spending.

That concentration can become dangerous when positions are financed with borrowed money.

A leveraged investor does not need the underlying investment thesis to be wrong for losses to become destabilizing. A sharp enough decline can trigger margin requirements, forcing the investor to sell assets regardless of its longer-term view.

That selling can then put additional pressure on the same securities held by other leveraged investors.

Situational Awareness provides a particularly visible example because of the size of its positions and the reported scale of its leverage. The fund’s rapid contraction also occurred during a period when investors were already questioning the enormous valuations attached to some AI-related companies and the ability of the industry to generate returns sufficient to justify its capital spending.

The SEC’s interest in the banks could therefore extend beyond the individual fund’s losses. Regulators may also be examining how prime brokers assess and manage counterparty exposure when multiple clients are pursuing similar AI-related strategies.

The issue is particularly relevant for stocks such as SK Hynix and CoreWeave, where strong institutional demand and concentrated positioning can contribute to rapid price movements in both directions.

A forced liquidation by one major investor can create losses for others even when the underlying companies’ business fundamentals have not materially changed. The recovery in SK Hynix and CoreWeave since the liquidation reveals that. The forced selling reflected the fund’s financing constraints rather than necessarily a fundamental reassessment of either company’s long-term prospects.

A fund generating strong gains can appear successful while taking on substantial hidden risks through borrowing, derivatives, and concentrated positions. Those risks become visible when markets move sharply against the strategy.

The SEC’s subpoenas could provide regulators with a clearer picture of how those risks accumulated around Situational Awareness and whether the fund’s prime brokers had sufficient visibility into its exposures.

The immediate investigation remains fact-finding rather than an enforcement proceeding. But the episode has already become a case study in the financial risks accompanying the AI boom, showing how a sharp reversal in a concentrated technology trade can rapidly turn into a liquidity and leverage problem when investors are heavily dependent on borrowed capital.

BlackBerry is seeing rapid growth in robotics within its QNX software business as the company seeks to position itself for the next wave of “physical AI,” CEO John Giamatteo told CNBC.

The shift marks another stage in BlackBerry’s transformation from a once-dominant smartphone manufacturer into an enterprise software company focused heavily on embedded and safety-critical systems.

QNX, BlackBerry’s largest software business, provides operating systems and embedded software used in safety-critical applications, particularly in the automotive industry. Its technology supports functions ranging from braking systems to advanced driver-assistance and semi-autonomous driving features.

BlackBerry says QNX software is now embedded in about 275 million vehicles, giving the company a large installed base from which to expand into other industries where software reliability and safety are critical.

Giamatteo said robotics is now among the fastest-growing areas within the QNX portfolio.

“We think that could be even a faster-growing segment of the industry for things like robotics,” he told CNBC’s “The Tech Download” podcast.

BlackBerry is not primarily targeting humanoid robots, which have attracted significant attention from Chinese and U.S. technology companies. Instead, the company sees a larger near-term opportunity in industrial and specialized machines, including warehouse robots, autonomous forklifts and medical equipment.

“As excited as we are about the dynamics of the automotive industry and where that’s going, these other applications around robotics and medical instruments and industrial automation represent a tremendous growth opportunity,” Giamatteo said.

“And one that I think BlackBerry and our QNX portfolio is really well positioned to address,” he added.

The opportunity is huge because modern robots increasingly require the same combination of computing, sensors, software and safety controls found in advanced vehicles.

Traditional industrial robots were generally designed to perform narrowly defined tasks in controlled environments. Newer systems can incorporate artificial intelligence to perceive their surroundings, interpret changing conditions and make decisions in real time. That shift is helping drive the concept of physical AI, in which AI systems move beyond generating digital outputs and interact directly with the physical world.

QNX provides a potential bridge between those two worlds for BlackBerry. The company already has years of experience developing software that must operate reliably inside vehicles, where system failures can have serious safety consequences.

Giamatteo said robotics is already one of the “fastest-growing businesses inside the QNX portfolio,” although BlackBerry has not disclosed how much revenue currently comes from the sector. The company has a QNX order backlog worth about $950 million, he said, with a portion of that backlog coming from robotics. Giamatteo did not provide a specific figure for the robotics component.

The backlog offers a more concrete indication of the opportunity than the company’s broader forecasts. While robotics remains a smaller business than automotive software, orders already in the pipeline suggest that QNX is being adopted for applications beyond conventional vehicle systems.

BlackBerry is also strengthening its position through partnerships with major chipmakers.

In April, the company announced an expanded partnership with Nvidia to deploy QNX software alongside Nvidia’s computing platforms in robotics, medical technology and industrial applications.

The partnership is essential because Nvidia has become a major supplier of the computing infrastructure used to train and run AI systems. Combining Nvidia’s AI computing capabilities with QNX’s embedded and safety-critical software could allow manufacturers to build robots capable of running sophisticated AI models while maintaining the reliability required for industrial and medical applications.

The strategy also gives BlackBerry exposure to robotics without having to manufacture robots itself.

Building physical robots requires substantial investment in hardware, manufacturing, supply chains and distribution. This means that software providers can potentially participate in the expansion of the industry across multiple manufacturers and applications.

BlackBerry’s automotive experience provides another advantage. Vehicle manufacturers have increasingly adopted software-defined architectures in which computing and software control a growing range of vehicle functions. Robotics is moving in a similar direction, with software becoming more central to perception, navigation, control and safety.

The company’s stock has roughly doubled this year as margins and profitability have improved, giving investors another reason to focus on whether its emerging businesses can supplement the established automotive operation. The challenge is that robotics remains a highly competitive market. BlackBerry faces established industrial automation companies as well as semiconductor and AI firms seeking to provide the computing and software layers for next-generation machines.

The company’s strongest argument is not that it can build the most advanced robot, but that QNX can become part of the underlying software infrastructure required to make robots safe, reliable and commercially deployable.

Analysts expect that to give BlackBerry a role in a market extending well beyond humanoid robots. Warehouses, factories, hospitals and other industrial environments may adopt specialized autonomous machines at a faster pace than consumers adopt general-purpose humanoids.

“Robotics is going to shift in a hardcore way … and we couldn’t be more excited about the opportunity in front of us,” Giamatteo said.

Billionaire investor Stanley Druckenmiller has sharply criticized the U.S. Treasury’s expanded bond-buyback programme, warning that attempts to influence long-term yields could undermine the credibility of the world’s most important government bond market while failing to address the country’s underlying fiscal problems.

Druckenmiller, a former colleague of Treasury Secretary Scott Bessent at Soros Fund Management, said investors were justified in interpreting last week’s decision to double the minimum size of Treasury’s long-end buybacks to $4 billion as an attempt at “price management” and called the move “a mistake.”

His comments, published Monday in the Wall Street Journal, add to growing criticism of the Treasury’s strategy as investors debate whether government intervention can meaningfully lower long-term borrowing costs without tackling the U.S. fiscal deficit.

The Treasury announced the larger purchases on Wednesday after the 30-year Treasury yield approached a nearly 20-year high. The announcement initially triggered a rally in government bonds and lifted stocks, but the gains quickly reversed as investors questioned the programme’s ability to address the forces pushing yields higher.

Druckenmiller stated that the 30-year Treasury yield is too important for policymakers to appear to be targeting it directly.

“The long bond yield is the most important price in the world,” he wrote, warning that intervention could create a cycle in which Treasury is forced to conduct larger purchases to defend a particular yield level.

That, he said, could damage one of the Treasury market’s most valuable characteristics: its reputation for predictable and reliable government financing.

“Debt management that even appears to follow the political calendar spends the one asset that took two centuries to accumulate: the credibility of the Treasury market,” Druckenmiller wrote.

“That asset doesn’t regain its value so easily.”

The timing of the buybacks is considered sensitive because the enlarged operations will run through the final months before the U.S. midterm elections.

Druckenmiller warned that even the appearance of a relationship between debt management and the political calendar could raise questions about whether Treasury is prioritizing market stability over long-term fiscal discipline.

The dispute comes as long-term Treasury yields have risen sharply this year, increasing pressure on U.S. equities and raising concerns about the cost of financing the government’s enormous debt.

Treasury Secretary Bessent has described the expanded purchases as part of a broader effort to improve Treasury-market functioning and influence the supply and composition of longer-dated debt. The department has said it will purchase more off-the-run securities, which are older Treasury bonds that are generally less liquid than newly issued benchmark securities. The purchases are intended to improve market liquidity and help manage the Treasury’s debt portfolio.

But Druckenmiller believes that the programme cannot resolve the fundamental problem confronting the bond market.

“You can’t buy your way out of a solvency conversation with liquidity tools,” he said.

His criticism underpins the distinction between debt management and fiscal policy. Treasury can change the maturity structure of government borrowing, conduct buybacks and alter the timing of issuance, but those measures do not eliminate the government’s underlying budget deficit.

Persistent primary deficits mean the U.S. must continue borrowing even if Treasury succeeds in reducing yields temporarily.

Druckenmiller said the government should instead allow the market to determine the appropriate level of long-term yields and concentrate on reducing the primary deficit.

“What should happen instead is straightforward,” he wrote. “Return buybacks to their stated purpose: small, scheduled.”

He argued that even a 5.5% yield on the 30-year Treasury should not necessarily be treated as a market crisis.

“If the 30-year must trade at 5.5% to clear, that isn’t a crisis. It is an invoice,” he wrote.

“Then do the only thing that durably lowers long-term yields: address the primary deficit.”

The comments come as an addition to the growing divide over how Washington should respond to rising borrowing costs.

Supporters of Treasury buybacks have noted that the government has a legitimate role in managing the structure and liquidity of the Treasury market. By purchasing older securities and potentially financing them with shorter-term debt, Treasury can alter the supply of securities available to investors at different points on the yield curve.

The approach can also improve market functioning during periods of thin liquidity or unusual price dislocations. But critics believe that such operations have limited power when yields are being driven by fundamental concerns about inflation, government borrowing and the supply of Treasury securities.

This matters because the U.S. government faces historically large financing requirements. If investors demand higher yields to absorb the growing volume of government debt, Treasury’s ability to buy bonds with relatively small amounts of cash may have only a limited effect on the overall market.

Druckenmiller’s warning also raises the issue of credibility. The Treasury market serves as the benchmark for pricing a vast range of assets globally, from corporate bonds and mortgages to derivatives and equities. U.S. government securities are also central to the international financial system because they are widely treated as a highly liquid reserve asset.

Any perception that Treasury is attempting to manage yields for political or short-term financial reasons could therefore have consequences beyond the immediate cost of government borrowing.

Druckenmiller is a notable critic because of his longstanding experience in global macro investing. He was a key architect of George Soros’ successful 1992 bet against the British pound and has worked alongside both Bessent and Federal Reserve Chair Kevin Warsh during his career in finance.

His argument also places fiscal reform at the center of the debate over long-term interest rates. If investors believe the government will continue running large primary deficits, Treasury may have to offer higher yields to attract sufficient demand. Higher yields then increase the government’s interest expense, potentially adding further pressure to future budgets. That creates a feedback loop in which rising interest costs contribute to larger deficits, requiring additional borrowing and potentially putting further upward pressure on yields.

Buybacks can influence the market’s plumbing, but they cannot break that cycle on their own.

The debate has become more urgent as the 30-year Treasury yield has moved above 5%, while the 10-year yield has also risen sharply. Higher long-term rates are already affecting equity valuations, corporate borrowing costs and investment decisions.

Investors see Druckenmiller’s warning as an indication that the market should focus less on whether Treasury can temporarily push yields lower and more on whether Washington can establish a credible path for reducing the primary deficit.

His central argument is that the bond market should be allowed to price the government’s fiscal position rather than be managed through increasingly aggressive purchases.

The Treasury’s next challenge will be to demonstrate that its buyback programme is primarily a debt-management and market-liquidity tool rather than an attempt to suppress borrowing costs.