- Vesta’s $30 million Series B, closed October 9, 2026, included three of its own lender customers as investors, putting originator capital directly behind the AI infrastructure they are running.
- Pennymac is targeting automation of 80% of identified origination tasks by end of 2027 using Vesta’s AI agents, with conditional approval times falling by 40% to 80%.
- A September 28, 2026 survey by AARMR, MBA and BCG found 59% of lenders cite regulatory uncertainty as the primary barrier to AI adoption, while many smaller institutions have no real-time bias monitoring in place.
Three of Vesta’s own mortgage lender customers put money into its $30 million Series B, closed October 9, 2026. That structure tells you something about where the industry is: originators are not just buying AI infrastructure, they are financing it. The gap the investment exposes is equally telling, speed gains are arriving faster than the compliance architecture needed to govern them.
Cutting Approval Times
Closing a loan typically takes around 40 days, largely because of manual labour across document collection, income verification and underwriting. Pennymac is using Vesta’s AI agents to automate tasks across the loan lifecycle, targeting 80% of identified origination tasks by end of 2027, with conditional approval times already down 40% to 80%.
Plaid moved in a similar direction on October 7, 2026, releasing new AI models aimed at improving lending decisions. Its LendScore 2 core model is, according to the company, 42% more effective at predicting repayment ability than traditional credit data alone. For home lending specifically, the Ls2 Home Lending model has enabled approval of 6.3% more borrowers at the same risk threshold, Plaid claims. Its Instant Link tool allows borrowers to share cash flow data for future applications in under two seconds.
WFG National Title expanded its DecisionPoint title automation platform in late 2026 to generate “touchless” title commitments for qualifying refinance transactions, integrating proprietary business rules with title data to automate underwriting without human review of each file. Other vendors offer comparable tools: Ocrolus, which the company says cuts income analysis time from 45 minutes to roughly 10, and Authentive, which uses agentic AI to gather employment and income data directly from source systems.
The Monitoring Gap
Speed gains carry compliance exposure. A survey released September 28, 2026 by the American Association of Residential Mortgage Regulators (AARMR), the Mortgage Bankers Association (MBA) and Boston Consulting Group (BCG) found that 59% of lenders cite regulatory and compliance uncertainty as the leading obstacle to scaling AI, ahead of ROI concerns, model reliability, data quality and talent constraints.
The Consumer Financial Protection Bureau requires specific, per-application reason codes for every AI-driven adverse action under Regulation B, generic checklists no longer satisfy the requirement. Algorithmic bias is the underlying concern: models trained on historical data can encode discriminatory patterns through proxy variables rather than explicit intent. That means lenders need continuous bias monitoring and documented variable justification, not just a policy on paper. Compounding the problem, many professionals use AI tools outside their organisation’s approved environments, creating compliance blind spots that written policies cannot cover.
Regulators Update the Rulebook
Federal banking regulators have also updated their supervisory approach. In April 2026, the Federal Reserve, the Office of the Comptroller of the Currency and the Federal Deposit Insurance Corporation issued SR 26-2, replacing the older SR 11-7 framework. The new guidance narrows what qualifies as a “model” under formal scrutiny for traditional banking operations, while signalling that separate guidance for generative AI and agentic AI is forthcoming. That gap leaves lenders in an interim period where investor requirements are filling the void: Fannie Mae’s Lender Letter LL-2026-04, effective August 6, 2026, requires seller/servicers to maintain written AI/ML governance policies with annual reviews and equivalent standards flowing down to vendors. Freddie Mac’s parallel requirements took effect March 3, 2026.
What Smaller Lenders Must Do Now
AI tool usage among lenders rose from 15% in 2023 to 38% in 2024, according to a 2025 Stratmor Group survey. The compliance infrastructure has not kept pace, particularly at smaller institutions.
Audit trail infrastructure covering inputs, model versions, timestamps and human overrides needs to be built in from deployment, not retrofitted later, it is the operational record regulators and GSEs will examine. The AARMR/MBA/BCG data makes the most visible gap plain: continuous monitoring for bias and model drift, moving from periodic review to real-time instrumentation, is what most smaller lenders have not yet built. Lenders also carry direct responsibility for AI tools their vendors deploy on their behalf, making third-party model and data due diligence a compliance requirement rather than optional vendor management.
The governance pressures building around agentic AI are not unique to mortgage lending, but the sector’s combination of high individual transaction values, fair lending law and GSE oversight makes the compliance stakes unusually concrete. Lenders operating across multiple states face a patchwork of state-level rules advancing even as federal frameworks remain incomplete, and written policies alone will not resolve it.



