AI on LP led secondaries – is the market ready?

Information and Data

Larger LPs typically have the models and the manpower to digest pages of LPAs, side letters, and GP reports. AI levels that playing field. Smaller LPs with less resource can now run diligence at a speed and depth that previously required a team of analysts. Transfer restrictions, fee waterfalls, ESG obligations and hidden economics can be surfaced in minutes.

This shift has the potential to rebalance information dynamics across the market, but it also raises questions about accuracy, oversight, and the standard of care expected from LPs using these tools. AI can surface issues quickly, but speed does not guarantee correctness. Models may misinterpret complex LPA provisions, overlook nuances in side letters, or generate false positives that require human judgment to contextualize. As a result, LPs will need to consider how much reliance can reasonably be placed on automated outputs and what level of human review and oversight remains necessary.

As these tools become more common, they may shape what constitutes a “reasonable” diligence process. LPs who choose not to use AI may need to justify why traditional methods were sufficient, while those who do use AI may need to show that they reviewed and validated the outputs rather than relying on them mechanically. In short, AI has the potential to enhance information quality, but it also introduces new expectations around oversight and decision making.

LP Portfolio Sales

Large portfolio sales are utilized where LPs seek to rebalance across vintages, sectors and geographies in a single transaction. These deals can involve dozens of fund positions, multiple GPs and a wide range of side letter rights and transfer mechanics. 

By analyzing large volumes of fund level and investor level data, AI tools can track fund performance in real-time, cluster assets with similar characteristics, identify outliers, and highlight positions that may require bespoke treatment. This can help LPs structure portfolios more effectively before going to market, improving both pricing outcomes and execution certainty.
Portfolio sales raise several challenges that AI may help address:

• Varied fund documentation: Each fund position will be governed by a different LPA, with its own transfer restrictions, consent requirements, and confidentiality provisions. AI can map these variations quickly and flag funds that may slow execution (e.g., due to advance notice requirements, ROFOs, delivery of legal opinions, etc.).
• Side letter complexity: Large LPs often hold bespoke rights across their portfolio. AI can help identify which rights are transferable, which are not, and where MFN clauses may be triggered.
• Pricing dispersion: Not all positions in a portfolio trade at the same discount. AI driven valuation tools can help LPs understand which assets drive value and which may require carve outs or alternative structures.
• Execution risk: With multiple GPs involved, the risk of delays or conditional consents increases. Predictive models may help LPs anticipate where bottlenecks are likely to occur.
• Process automation: Portfolio sales typically involve extensive coordination across multiple stakeholders and workstreams. AI allows labor intensive aspects of project management to be automated, reducing costs and minimizing human error. 
• Legal negotiations: AI can streamline legal work by flagging issues for consideration, suggesting fall-back positions based on a predefined playbook and assisting in drafting rider language based on precedent transactions. This improves efficiency and ensures consistency in the commercial outcome across the entire portfolio.  

AI does not eliminate the legal and operational complexity of portfolio sales, but it can help LPs navigate that complexity more efficiently and with greater confidence.

Pricing

AI driven valuation engines can triangulate GP reported NAVs with historical performance, macroeconomic data, and comparable trades. In an auction process, price discovery also has the potential to become quicker and more transparent. From a legal perspective, this introduces a new challenge: if AI can produce a defensible fair value range, does relying on anything less become a liability?

LPs are expected to take reasonable steps to understand the value of what they are selling and to avoid obvious errors or omissions. If AI based valuation tools become widely used and are shown to improve pricing accuracy or highlight material discrepancies, it may become harder to justify ignoring them entirely. 

That does not mean LPs must accept AI outputs at face value. The duty is to consider the information available, not to follow an algorithm blindly. A well governed process might involve comparing AI generated ranges with GP reported NAVs, market comparables and advisor input, and documenting why certain methodologies were preferred.

In practice, the liability risk arises not from declining to use AI, but from failing to demonstrate a thoughtful, well supported valuation process. As AI becomes more common, LPs who can show that they evaluated its outputs may be better positioned to defend their pricing decisions.

Fiduciary Duties

When an LP sells a stake, they are expected to act prudently and in the best interests of their beneficiaries. AI complicates this analysis. If an AI model identifies risks or opportunities that humans miss, does ignoring those insights breach fiduciary duty? Conversely, does relying on AI create a new duty to understand and validate the model’s assumptions? 

Fiduciary duties require LPs to act prudently, to make informed decisions, and to exercise appropriate oversight. AI does not change those duties, but it may influence what “prudence” looks like in practice.

If AI tools become widely adopted and are shown to improve the quality of diligence or pricing analysis, a court or regulator could view them as part of a reasonable decision making process. In that scenario, ignoring material AI generated insights (particularly where they highlight clear risks) could be difficult to justify. Conversely, LPs could point to such AI insights (supported by advisor inputs) as evidence that they have discharged their fiduciary duties. In either case, documenting the rationale for reliance upon, or rejection of, AI-created outputs will be prudent. 

At the same time, reliance on AI does not eliminate the need for human judgment. If an LP chooses to use AI, they may be expected to understand the tool’s limitations, data sources, and potential biases. This implies a duty to ensure that the model is appropriate for the task.

The emerging consensus is that AI does not create new fiduciary duties but it does raise the standard for what counts as a well informed, well documented process. LPs who integrate AI thoughtfully, and who can demonstrate how they evaluated and contextualized its outputs, will be better positioned to show that they acted prudently.

Confidentiality

LPAs were never drafted with AI in mind. Feeding fund level confidential data into third party AI systems therefore raises several legal questions:

• Are LPs inadvertently breaching confidentiality clauses? Most LPAs restrict disclosure of fund information to third parties except in defined circumstances. Whether an AI provider qualifies as a permitted disclosee depends on the contract. LPs may need to ensure that vendor agreements impose confidentiality protections at least as robust as those in the fund documents. 

• Can AI vendors train models on sensitive fund information? Many AI tools rely on model training or data aggregation. Unless expressly permitted by GPs, this may constitute a disclosure to unknown third parties or embed sensitive information in ways that are difficult to control. In particular, GPs may be concerned that sensitive pricing information could become widely available in the market, adversely affecting their successor fundraise. LPs will likely need contractual assurances that fund data is used only for their own purposes and is not incorporated into shared models. 

• Who is liable if data leaks through an algorithmic back door? Liability will turn on the vendor contract. LPAs typically place responsibility on LPs for maintaining confidentiality, meaning LPs may remain accountable to the GP even if the vendor suffers a breach. Vendor agreements therefore need to address security standards, audit rights and indemnities. A “Zero Data Retention” policy may be the most prudent approach when selecting AI vendors.

These issues demonstrate that AI does not change the underlying confidentiality obligations, but it does make compliance more complex. LPs, GPs and advisors will need to review both fund documents and vendor contracts to ensure that AI driven workflows fit within existing legal frameworks.

Conclusion

AI is beginning to influence LP led secondaries in ways that touch pricing, diligence, confidentiality and process. It does not replace established legal frameworks, but it does interact with them in ways that market participants will increasingly need to navigate. The core duties (e.g. acting prudently, protecting confidential information and ensuring fair processes) remain unchanged. What may evolve is how those duties are interpreted as AI tools become more common and more capable.

As the technology matures, the market will develop a clearer sense of what “good practice” looks like. Until then, a balanced, governance led approach offers the most reliable path forward.

Authored by Leanne Moezi, Thomas Kim, and Adam Brown.