Continuation vehicles in the crosshairs

Topline

• SEC turns its eye toward PE continuation vehicles

• Anthropic export controls could shape the future of AI regulation 

• House Democrats press SEC on consumer-facing AI trading agents

• Industry seeks to head off standalone private credit reporting

• Quick hits

SEC turns its eye toward PE continuation vehicles

The SEC is reportedly examining private fund continuation vehicles, scrutinizing a structure that has boomed as exits have slowed and private equity managers sit on a growing backlog of unsold portfolio companies. Continuation vehicles allow managers to move assets from older funds into newly created vehicles, giving existing investors a liquidity option while extending the holding period for assets that may be difficult to sell in a sluggish exit environment. These structures give managers more time to hold companies while returning capital to investors without forcing a sale at a depressed valuation, selling to a competitor, or realizing potential losses under current market conditions.

Continuation vehicles have become an important pressure valve for private equity. But the same features that make continuation vehicles useful have also made them a regulatory flashpoint. CVs are inherently conflicted because managers are often on both sides of the transaction: selling assets out of an existing fund while continuing to manage them in a new vehicle. Instead of pursuing new requirements as the now-vacated private fund advisers rule sought to do with mandatory fairness opinions, the agency is relying on existing fiduciary obligations, disclosure standards, and anti-fraud authority to test whether fund managers can justify how these transactions are structured, priced, and disclosed. 

Why it matters: Much of the public framing around the SEC’s private market scrutiny has been directed toward private credit and retailization, including the risks that arise as illiquid private fund strategies reach retirement plans, interval funds, and other products marketed to individual investors. The recent attention on continuation vehicles shows that the SEC’s heightened regulatory focus extends beyond retail-facing products (as we have previously highlighted) and into institutional market structures that have become central to how private equity manages liquidity. This aligns with the agency’s 2026 exam priorities, which flagged valuation, fees, conflicts, and differential investor treatment across private fund structures as areas of focus.

The practical takeaway: Process matters. As scrutiny rises, fund managers will need to show that continuation vehicle transactions are well-governed, well-documented, supported by credible third-party valuation work, and defensible as fair to investors cashing out, rolling over, or buying into the fund. 

Anthropic export controls could shape the future of AI regulation

The Trump administration’s intervention in Anthropic’s Fable 5 and Mythos 5—and now OpenAI’s GPT-5.6 rollout—is quickly becoming a test case for the next phase of AI regulation. The immediate story is about a disputed jailbreak, a Commerce Department directive, and a restricted model launch. The bigger story is that Washington appears to be moving from abstract debates over frontier AI oversight toward a government-mediated launch regime for the most advanced models.

What happened: Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9. Fable 5 was made broadly available, while Mythos 5 was released through a narrower access program focused on advanced cybersecurity users. Three days later, the Commerce Department directed Anthropic to suspend access to both models by foreign nationals, including foreign-national employees inside the United States. Because Anthropic could not reliably screen all users by nationality in real time, the company took the models offline more broadly.

The directive followed concerns that Fable’s safeguards could be bypassed in ways that exposed cyber-related capabilities associated with Mythos. Anthropic has pushed back, arguing that the reported jailbreak was narrow, non-universal, and not meaningfully different from vulnerabilities that affect other advanced models not subject to similar controls. That distinction matters. If the standard is that a frontier model must be immune from all jailbreaks before release, no major model is likely to satisfy it. If the standard is tied to measurable capability thresholds, mitigation practices, access controls, monitoring, and response plans, developers and regulators have a clearer basis for decision-making. Two weeks later, the models remain offline as the two sides have yet to reach a resolution.

EO tension and legal uncertainty: The episode exposes a central tension in the administration’s AI policy. Earlier this month, the White House issued an executive order creating a voluntary 30-day pre-release review process for advanced AI models, while expressly disclaiming any mandatory licensing, preclearance, or permitting regime. But the Anthropic directive shows how voluntary review, export control authority, and national security concerns can quickly converge into something that looks like de facto launch approval.

A voluntary review process may remain voluntary on paper, but once the government demonstrates it can pull a model from market days after release, pre-release engagement starts to look less optional. Developers may conclude that documented review, government testing, and a clear mitigation record are not merely good safety practices; they are market-access protection. That dynamic is already beginning to shape launch strategy across the frontier AI ecosystem, as companies weigh whether to stagger releases, limit access to approved customers, or seek more explicit federal engagement before broad deployment. This is apparent with OpenAI’s forthcoming GPT-5.6 model, which will initially be available only to a small group of U.S. companies and organizations approved by the administration. The company reportedly did not plan to restrict the general-use model, but shifted to a controlled release after White House engagement.

The legal hook is also unsettled. Export controls traditionally focus on sensitive technology, technical data, software, source code, chips, computer infrastructure, and transfers to foreign parties. Hosted model access is different. Users may receive outputs, but not the underlying model weights or source code. That raises a threshold question of when does access to an AI service become an export of controlled technology. Separately, Legion LegalTech sued the government after losing access to the models, arguing that the restriction exceeds Commerce’s authority because hosted model access is not the same as transferring software, source code, or model weights.

Congressional action: In response, a bipartisan group of House lawmakers sent Commerce Secretary Howard Lutnick a letter demanding the legal basis, technical evidence, evidentiary threshold, and restoration criteria behind the Anthropic directive. 

Congress is unlikely to resolve the broader AI debate quickly, but the lack of clear standards and increasing concern around advanced models have created more urgency for legislative action. The episode shows the danger of regulating frontier AI through opaque directives instead of clear statutes, transparent standards, and defined appeal rights.

Why it matters: This is the beginning of a new AI regulatory perimeter. The debate is no longer just whether Washington should regulate frontier models; the question is how. 

For end users, AI dependency is now a policy, compliance, and operational risk issue. Companies using frontier models for software development, cybersecurity, legal work, financial analysis, valuation, portfolio monitoring, or customer workflows should assume that access to specific models can change quickly based on national-security determinations, vendor controls, or government review. The firms best positioned for it will treat model governance like vendor risk management and build compliance processes that can adapt as federal standards evolve.

House Democrats press SEC on consumer-facing AI trading agents

House Financial Services Committee Democrats sent a letter to SEC Chairman Paul Atkins on agentic AI retail trading, specifically asking what oversight, guidance, and investor protections are in place as brokerage firms begin allowing AI agents to make autonomous trades on behalf of retail investors—and whether existing securities laws are sufficient or Congress needs to act. Third-party AI agents are beginning to operate inside registered retail platforms, while platform disclosures often warn that the tools carry significant risk and may not be controlled, supervised, monitored, recommended, or audited by the brokerage. Democrats argue that disclaimers alone do not resolve the allocation of legal responsibility among broker-dealers, AI developers, and retail investors when AI agents make or enable consequential investment decisions in retail accounts.

The inquiry goes directly to the regulatory perimeter. The lawmakers ask when an AI agent or developer would need to register as a broker, dealer, investment adviser, or associated person; how third-party AI agents affect broker-dealer obligations under Regulation Best Interest (Reg BI), supervision, books and records, best execution, privacy, cybersecurity, and order-handling rules; and whether AI developers bear liability when their tools facilitate autonomous retail trading. They also press the SEC on market-structure risks, including herding, volatility, hallucinated or manipulated information, prompt-injection attacks, social media rumors, and misuse of customer trading data.

Why it matters: As agentic AI becomes more embedded in financial services, calls for regulatory clarity will only increase. Financial markets are built on regulatory frameworks that assume identifiable actors at each consequential step: the adviser who recommends, the broker who executes, the intermediary that supervises, and the firm that bears legal responsibility. Agentic AI complicates that structure by inserting software that can analyze, recommend, and execute with limited or no human involvement. Reg BI, the Investment Advisers Act, and broker-dealer registration rules were not written with autonomous retail trading agents in mind. The SEC’s response will be an early test of whether existing law can stretch to cover agentic finance, or whether Congress will need to define new rules for AI systems operating inside regulated markets.

Industry seeks to head off standalone private credit reporting

A coalition of major private credit trade groups is urging the SEC and CFTC not to build a standalone private credit reporting regime into Form PF. In a joint letter, the industry argued that existing Form PF requirements already capture the core data regulators need to monitor private credit funds, including leverage, credit quality, counterparty interconnections, investor liquidity, and portfolio composition. Rather than create a new private credit-specific reporting architecture, the groups are pressing for targeted enhancements to existing Form PF sections, including modest additions around borrower defaults, weighted average loan life, concentration in top positions, and industry exposure.

Why it matters: Private credit remains squarely in regulators’ line of sight, but the debate is shifting from whether the market should be monitored to how granular, standardized, and costly that monitoring should be. The industry contends that regulators can get better visibility into leverage, liquidity, credit quality, and concentration without forcing advisers to build a duplicative compliance infrastructure for a market that policymakers have said does not currently present systemic risk concerns. The SEC is unlikely to abandon private credit monitoring altogether, but it may be receptive to the argument that the final rule should refine Form PF rather than create a wholesale private-credit reporting overhaul.

The bottom line: The direction of travel is more standardized visibility into private credit risk, but likely through targeted Form PF refinement instead of a standalone regime.

Quick hits

• House and Senate advance bipartisan housing bill. Congress sent a bipartisan housing reform bill to the president’s desk after months of negotiation and revision, clearing one of the Senate’s major financial services priorities from the calendar. The development is significant in its own right, but it also improves the scheduling posture for the INVEST Act—the capital formation package that passed the House in December and has been awaiting Senate consideration. With housing off the table, Senate Banking has more bandwidth to turn to other priorities, including capital formation, though crypto market structure remains ahead of INVEST in the queue. The bigger constraint is time: the legislative window is narrowing quickly before August recess and the midterm sprint this fall.
  

• Trump seeks to boost quantum computing with new executive orders. President Trump signed two executive orders on quantum technology, pairing an industrial-policy push to win the quantum race with a faster transition to quantum-resistant cybersecurity. The first order establishes a whole-of-government strategy for quantum information science and technology, including a new effort to deliver a scientific-scale quantum computer to a Department of Energy facility, strengthen domestic supply chains and workforce pipelines, protect sensitive technology from adversaries, and accelerate quantum sensing and networking. The companion order is the more immediate compliance signal. It directs federal agencies to migrate high-value and high-impact systems to NIST-approved post-quantum cryptography for key establishment by the end of 2030 and digital signatures by the end of 2031, while helping critical infrastructure operators plan their own transitions. The point is not that quantum computers will break encryption tomorrow, but that sensitive data stolen today may still be valuable when quantum decryption becomes possible later. For financial firms and private market participants, the EO is a signal to begin inventorying cryptographic systems, reviewing vendor exposure, and building post-quantum migration plans now.

• Trump kills plan to quickly confirm new intelligence chief. Former SEC Chairman Jay Clayton’s path to director of national intelligence hit an unexpected obstacle after President Trump canceled his Senate Intelligence Committee hearing hours before it was set to begin, saying the nomination would not move forward until James McDonald is confirmed to replace Clayton as U.S. attorney for the Southern District of New York. Bill Pulte, whose acting DNI appointment had already triggered Democratic opposition and complicated FISA reauthorization talks, will remain in the role for now—and he’s already making significant changes while in the role, which could further complicate the confirmation process. With FISA lapsed and no confirmed intelligence chief, the prolonged stalling carries real national security costs and further complicates the Senate’s ability to move on other legislative priorities.

• Bessent and Lutnick clash over government AI stakes as industry pushes back. Senior Trump administration officials are debating whether, and how, the federal government should acquire equity stakes in major AI companies, with Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick on opposing sides. While Bessent champions directing any AI equity toward Trump Accounts, Lutnick favors a sovereign wealth fund structure. No decision has been made, and industry reception has been cold. Some in the industry have publicly rejected the concept, and policymakers have expressed skepticism. Structural problems abound, as most major AI companies don't pay dividends, meaning any government stake would appreciate on paper with no clear return mechanism. Deeper concerns around regulatory capture persist as well: A government holding a financial position in Anthropic or OpenAI would have a built-in incentive to soften safety oversight. 

• Newsom calls for national billionaire tax while fighting California wealth tax measure: This morning, California Gov. Gavin Newsom unveiled a national wealth tax agenda framed as an "economic reset for America"—and a signal of a 2028 presidential run. The proposal would impose a minimum tax on anyone with a net worth above $100 million, restrict the practice of borrowing against appreciated assets as an alternative to realizing taxable income, tighten inheritance rules, raise corporate taxes to pre-Trump levels, and create a national public equity fund giving Americans a stake in the AI industry. The timing is pointed: Newsom rolled out the agenda one day after a California SEIU-backed ballot initiative imposing a state-level billionaire wealth tax qualified for the 2026 ballot—a measure he opposes, arguing that wealth is mobile and the fight belongs at the federal level. Newsom argues that distinction matters: wealth is mobile, and a state-level tax can be avoided by relocating, while a federal minimum tax cannot. The proposal isn’t expected to advance under the current Congress, but it adds another data point to the growing bipartisan pressure around AI equity and wealth concentration—and gives Newsom a platform to run on.

Sign up below to receive Carta's Policy Weekly Brief