BENEFITS DIGEST

Ary Rosenbaum opens with the Terminator and lands on a practical point. He expects artificial intelligence to compress fees and margins in plan administration, which would let providers who use it well do more for less while the ones who don’t lose business to the ones who do. He wants human review of anything the tools produce, controls on what client data goes into them, and attention to the conflicts that come with the efficiency.

Kathryn Moody’s weekly roundup leads with the collision between disability-accommodation law and paid-time-off policies, pointing readers to the publication’s fuller treatment of when leave itself becomes the accommodation. The rest of the week’s sweep runs from Allegheny County’s new paid-leave mandate to AI-generated busywork, mental health benefits and changes to how the federal employment-discrimination agency handles settlement talks.

The SEC’s examination priorities for 2027 put private-market access, conflicts of interest, cybersecurity and evolving technology on the supervisory agenda. Practitioners quoted in the article read the list as a signal for advisers serving retail investors, including retirement savers moving into less liquid investments. Examination priorities direct the staff’s attention and don’t create new duties by themselves. Advisers and the committees that hire them can use the list to anticipate what examiners will ask about.

Writing for benefits brokerage Borislow, Sena Meilleur argues that smaller employers can no longer treat renewal season as the moment to manage catastrophic claim risk. GLP-1 pharmacy spend, cell and gene therapies that can cost more than $1 million per claim, and AI-driven coding intensity are reshaping medical budgets, so the useful decisions, about self-funding, captives and reinsurance against the biggest claims, happen early in the plan year, before an adverse claim limits the options. She treats a good renewal as a signal to dig in rather than coast.

Preparing for a retirement-plan committee meeting can mean searching documents and reconciling information from several providers. The AI tools described in this feature aim to automate parts of that work, including document queries, meeting preparation and fiduciary reporting; other tools focus on participant engagement. These are vendor descriptions, not independent findings that the products improve outcomes. Before adopting them, sponsors need to establish what data the tool can access, who checks its output and who remains responsible for the resulting decisions.

An AI assistant may make an old access-control problem much easier to exploit. In Brittney Mollman’s hypothetical, the technology exposes information that users already had overly broad permission to reach. Blocking the assistant alone would leave those underlying permissions in place. Benefits teams handle medical, payroll and retirement information, making the allocation of responsibility especially important. Someone must own the access review, assess any exposure and verify that the permissions themselves have been repaired.

NAPA reports a practitioner panel with Jordan Cross, Amanda Iverson, and Joey Santos Jones. The panelists describe artificial intelligence already at work in preparing employee data for plan administration, drafting participant communications, and building training materials, and they map where human judgment still has to do the work. Their cautions are confidentiality, unreliable answers, and the slow loss of the skills needed to check a tool’s output, so they recommend written policies and human review. It is practitioner conversation rather than regulatory guidance, and delegation to a tool does not delegate the fiduciary duty.

K&L Gates explains that California employers with five or more employees must extend bereavement leave to a designated person beginning January 1, 2027, with employees able to designate one person per 12-month period for the five-day leave. The roundup also covers revised restrictions on stay-or-pay agreements, the arrangements requiring workers to repay bonuses, training costs, or other amounts when they leave, with separate conditions and effective dates, plus AI, surveillance, and arbitration changes. Effective dates differ by statute, so compliance calendars should use each law’s own date rather than a single January 1 assumption.

Cyber insurers at the NetDiligence summit say most losses still come from familiar ransomware and human error rather than exotic AI attacks, which keeps the basics decisive, multifactor authentication, network segmentation, employee training, and monitoring that someone actually watches. For benefits teams the relevance is direct, since plan data and vendor connections are exactly the kind of target these controls protect, and carriers increasingly price coverage on their presence.

An ASPPA Annual panel turned Cycle 4 plan restatements into a project plan. Identify which plans need restating, chase down missing or unsigned prior documents, decide up front whether clients may request design changes during the cycle, and run one status-tracking system with clear internal ownership. The panel’s sharpest advice is to carry current provisions forward accurately and put experienced document reviewers on the output, because restatement software and AI drafting do not replace human review. Batching the work beats a year-end scramble.

WTW’s global survey of 41,000 employees across 31 markets finds 70% saying their benefits meet their needs while 38% worry about AI harming job security, and reports that financial wellbeing has deteriorated even as other measures improved. The firm argues for more tailored support. The figures are global rather than U.S.-only, and the detailed report sits behind a registration form, so this introduction is what is freely readable.

Don Trone argues that AI can sharpen a fiduciary committee’s analysis without displacing the committee’s responsibility for judgment. His leadership, stewardship and governance framework treats generated answers as inputs that still need human oversight, relevant evidence and accountability for consequences. The piece is a practitioner’s opinion about sound process, not a Labor Department requirement or a litigation safe harbor, and the framework is Trone’s own.

EBN gathers practitioner advice on making workplaces and benefits more usable for neurodivergent employees. The practical questions for sponsors run to whether mental-health benefits and flexibility are structured so employees will claim them, how enrollment and accommodation processes handle employees who have not disclosed a condition, and whether AI tools in employment practices introduce bias. It draws on earlier reporting and practitioner interviews rather than a new study.

Covington’s roundup leads with New York’s personnel-records law taking effect November 8, which gives employers five business days to provide copies, requires retention for three years after employment ends, and adds a 10-day notice when negative information goes into a file. New York City’s rules add 32 hours of unpaid sick leave to the existing 40 or 56 paid hours plus 20 hours of prenatal leave, under a final implementing rule issued July 23. Severance, job-posting, employment-waiver and AI-reporting bills still await the governor and are not yet obligations.

EBRI and Greenwald Research surveyed 1,238 workers ages 21 to 64 in late June and July for the seventh annual edition of this survey. Concern eased slightly across every well-being measure, with financial worry at 6.0 on a 10-point scale, down from its 2022 high of 6.9, yet 70 percent of workers report problematic debt and 57 percent say they often or always feel sick at work. Workers pointed to retirement plans and health insurance as their leading sources of financial security, at 64 and 63 percent, while only 52 percent are very or extremely satisfied with their benefits package and 45 percent name a bigger employer contribution as the most valuable improvement. Two findings deserve a benefits manager’s attention. A quarter of insured workers now say their plan covers GLP-1 drugs for weight loss, up from 16 percent last year, and 45 percent agree employers should cover them. And on AI, 71 percent are concerned about bias in AI tools used for benefit recommendations, while given the choice only 15 percent would pick an AI-powered benefits counselor over a human one.

Fintech M1 has launched M1 Advisor, an artificial intelligence financial adviser built into its investing platform and free to clients through 2027. The company reactivated its registered investment adviser entity to offer non-discretionary advice with no human advisers behind it, running on models from Google, OpenAI, Anthropic and Amazon under an internal governance process. The RIA reports $1.85 billion under management, and the system can see a client’s outside 401(k)s, IRAs and loans through Plaid, though it only advises and cannot act. The claim worth watching from the plan world is the legal one, SEC-registered fiduciary advice generated entirely by machine. Participant-facing advice tools carry the same questions about conflicts, disclosures and accountability when the model is wrong, and somebody was going to run this experiment first.

Governor Newsom signed most of the session’s employment bills and vetoed AB 1940, which would have added perimenopause, menopause and related conditions to FEHA’s definition of sex; a companion trigger provision in AB 2563 stays dormant as a result. For benefits and leave practitioners the headline is SB 1149, which extends California’s five days of job-protected bereavement leave to one “designated person” of the employee’s choosing per 12-month period, beyond the current family definition. AB 1697’s fix to the stay-or-pay law, which pushes its restrictions to contracts signed on or after January 1, 2027 and adds exceptions, is already in effect. The AI cluster deserves attention on its own. SB 947 bars using an automated decision system as the sole basis for discipline or termination, AB 1883 bans emotion recognition and neural data collection in workplace surveillance, and SB 951 requires Cal/WARN notices to flag layoffs substantially caused by AI. Most provisions take effect January 1, 2027, with the automated-decision rules operative July 1, 2027.

TIAA’s “Retirement in the Age of AI and GLP-1s” survey polled 1,000 adults through KRC Research in late July. On longevity, 53 percent worry more about outliving their savings than about underspending, and 43 percent doubt traditional retirement planning accounts for lives that stretch into the 90s. On what AI will do to their finances, 27 percent expect breakthrough treatments to raise costs, 22 percent expect efficiency to lower them, 20 percent expect little effect, and the rest can’t say. Separately, a third are seriously concerned AI could threaten their earning potential before they retire, a worry that peaks with Generation Z. TIAA’s read is that these are all longevity problems, and its consultants’ advice to employers is to weigh guaranteed lifetime income in plan design on par with fees and the investment menu.

TIAA's survey of 1,000 adults finds 53 percent worry more about outliving their savings than about underspending in retirement, and 83 percent carry financial concerns about living longer. Younger respondents were more worried about AI. Forty-two percent of Generation Z respondents are extremely or very concerned AI will hurt their career and retirement savings, against 28 percent of Generation X and baby boomer respondents combined. TIAA, which sells annuities, argues that guaranteed retirement income can help address those concerns.

The Hill's read on the second Make America Healthy Again summit, where administration officials shared panels with insurers and corporate executives. Two discussions stood out for employers. Officials including Secretary Kennedy and CMS administrator Oz pressed for wider use of AI in healthcare, with Kennedy claiming AI can outperform a doctor's second opinion. And the summit's opening panel put psychedelic medicine developers alongside an HHS lawyer, weeks after the FDA took public comment on therapeutic uses of psychedelics. UnitedHealth, Elevance, and Walmart were among the participating companies, and The Hill reports some companies were offered access to officials for a fee.

Dairy Farmers of America closed its St. Albans Creamery in Vermont, the Teamsters local took the closure to arbitration as anti-union retaliation, and the company announced it would stop paying its 85% share of health premiums for the 44 bargaining-unit members on October 1. Judge William Sessions III ordered the premiums continued until arbitration concludes, with an exception for anyone who lands coverage with a new employer. The judge found that lost health coverage is the kind of harm an arbitrator can't later repair, and courts can order an employer to preserve the status quo so that arbitration doesn't become a hollow formality. Employee declarations described a $15 copay becoming a $3,000-a-month prescription and family COBRA premiums near $2,500. The court declined to force an arbitration date, since the parties' chosen arbitrator isn't available until January, and noted a disputed internal memo proposing 'idling operations' aimed at 'Union Decertification,' which the company says is an AI-generated fake.

More than three-quarters of calendar-year multiemployer pension plans surveyed by Segal are in the green zone for 2026, meaning they are not classified as endangered or critical, and counting plans receiving special financial assistance there are more well-funded plans now than in 2008, when zone certifications began. Two federal appeals courts also held this year, in Reichert v. Kellogg and Drummond v. Southern Company Services, that the assumptions used to calculate different pension payment options must be reasonable. Those assumptions can affect how much a retiree receives when choosing, for example, a pension that continues paying a surviving spouse. The courts did not prescribe a single acceptable method. The recap also carries quarter-in-review notes on the Federal Reserve and a short framework for using AI on plan documents and participant communications, with human review of everything AI produces.

A fight over who is inflating hospital bills, with plan dollars in the middle. The Blue Cross Blue Shield Association says AI-assisted coding tools added about $942 million to its costs over two years without a matching rise in treatment. It points to more anemia diagnoses without more transfusions. PwC analysts separately attribute 9 percent of next year's increase in insurer medical costs to AI billing tools. Billing companies, whose coding platforms reach most U.S. hospitals, say the tools identify patient conditions and treatment complexity that hospitals previously failed to document fully. Employers that pay their own health claims should ask their claims administrator how it checks higher charges generated by AI-assisted billing.

Quarles recommends preparing for California's five health-care AI bills together, with an inventory of AI tools and procedures for human oversight, disclosure, and consent. The measures address clinical decisions, bias in decision-support tools, protection for clinicians who override AI, psychotherapy, and access to human customer-service agents. The chatbot bill, AB 1609, still awaits the Governor's signature. Quarles urges organizations to prepare for potential January 1, 2027 requirements. Health plans and pharmacy benefit managers serving California members should review their chatbots and vendor contracts.

AI can help employers process leave requests more efficiently and consistently, writes Gallagher's Alec Herring in Employee Benefit News. But using the technology doesn't reduce an employer's responsibilities under the ADA or FMLA. Employers remain accountable for their leave decisions even when software handles much of the work.

A California bill on the Governor's desk, AB 1609, would give health care customers a right to a human being. Businesses with more than $500 million in annual revenue serving California customers, including health plans, payers, managed care organizations, and PBMs, would have to disclose when a chatbot isn't a person, connect a customer who asks about a refill, a transfer, a denied claim, a prior authorization, or a specialist referral to a live agent within 15 minutes or offer an appointment within one business day, and cap cumulative phone holds at an hour. The bill asks for a good-faith effort, not a stopwatch guarantee. Licensed hospitals won a broad carve-out; plans and PBMs didn't. Penalties reach $10,000 per violation, effective January 1, 2027 if signed. Sponsors with California populations should ask vendors now how member-service AI would meet the clock.

Bloomberg Opinion's Allison Schrager argues that with more than 60 percent of Americans owning stock, largely through $32 trillion in IRAs and DC plans, concentrated AI valuations have become a risk to household savings: “we are all betting on AI taking over the economy.”

A different question about AI, from Don Trone: the consequential one is no longer whether fiduciaries have the right information but whether they have “the moral and ethical capacity to act on it,” and AI, he argues, can supply the transparency that makes acting easier than rationalizing.

Segal urges health plan fiduciaries to include vendors' use of AI in their oversight, warning about the risks of deploying tools without adequate review or participant protections. Its read on the technology is sober. The most consistent measurable benefits so far are administrative, while generative chatbots, the most visible use, have been shown to enable harmful behaviors in vulnerable users, against what Segal calls the lack of a clear regulatory framework for AI tools in mental health. The practical takeaway is four steps: audit vendors' use of AI, require clear disclosure of participant-facing AI, adopt ongoing governance practices, and verify the safety guardrails. A fuller version with a risk table by AI type and questions to put to vendors runs in this quarter's Benefits Quarterly.

The premise is simple—automation does not transfer liability. The guide works through six exposure points: AI generated participant communications that still must satisfy disclosure rules, claims administration where a vendor's biased or inadequate algorithm becomes the hiring fiduciary's problem, investment menu tools that should supplement rather than replace a live 3(21) or 3(38), committee minutes that come out either dangerously verbatim or uselessly sparse, confidentiality and HIPAA risk when nonpublic data reaches AI features, and service agreements that should name AI uses, testing, and approval rights.

The sources converge on one design principle—keep a human in the loop and make access to a counselor a standing choice rather than an escalation. Gallagher's Robby White urges advisers to steer employers toward communications, decision support, and navigation first, inside a governance framework that is transparent about what data the tools use. Collective Health's Ali Diab says autonomous agents making healthcare decisions are not here yet, though the direction is set.

With 59% of Americans lacking the savings to cover an emergency expense, employers are rethinking financial wellness benefits, from emergency savings programs to retirement plans to AI-powered financial advice.

An opinion piece argues AI's best use in open enrollment is taking over the repetitive administrative work so benefits teams can spend their time actually helping people, citing research that only 69% of employees feel they understand their medical benefits. The pitch is efficiency in service of the human conversation, not a replacement for it.

With most states yet to enact AI-specific employment laws, the contract is the control. A checklist of provisions for employers deploying AI tools through HR vendors.

As artificial intelligence continues to move into the mainstream, opportunities abound for eliminating confusion, personalizing content and improving outcomes.

New NIRS survey shows deepening concerns about retirement viability in current economy, plus a distrust of AI financial advice and crypto investments.

The retirement industry may be overdue for a "reintroduction."

Nearly half of people with an ongoing mental health condition who use large language models say they lean on AI for mental health support, per a survey EBN reports, most commonly for anxiety and depression. For benefit leaders weighing AI tools in workforce mental health programs, the vetting question is no longer hypothetical.

The Prudent Investment Fiduciary Rules blog proposes a quantitative framework for documenting fiduciary prudence, combining cost-comparison metrics with AI-assisted process tools. Dense, but of interest to committees formalizing their prudence files.

Mercer's quarterly state survey: Virginia enacted both paid family and medical leave and paid sick and safe leave, several states passed major PBM laws including a Tennessee statute challenged in court the day it took effect, and Alabama, Georgia, and Texas legislated on AI use in benefits.

With salary budgets flat and AI reshaping rewards, employers lean on transparency, career growth, and benefits to retain talent.

Principal's latest financial-health index finds employers and employees converging on two trends: workers postponing retirement dates and both sides leaning on AI tools for planning.

Global retirement experts suggest investment options, advice, and artificial intelligence can all help with drawing down defined contribution savings, with lessons from decumulation designs abroad.

An NCPERS survey suggests lower-stakes experimentation with artificial intelligence is happening more quickly than strategic integration.

WTW 2026 Defined Contribution Survey report finds employers are asking defined contribution plans to do more than they were originally designed for, prompting greater focus on retirement income, AI, governance, and measurable participant outcomes.

Doing so may influence participant decision-making in ways that raise disclosure and accuracy concerns and, in certain cases, could be viewed as recommending action or inaction by participants.

Don Trone examines how to build new protocols for fiduciary governance.

Colorado employers face new compliance obligations under recently enacted laws addressing artificial intelligence-driven employment decisions, demographic workforce reporting, disability accommodations, wage-and-hour requirements, and Family and Medical Leave Insurance administration, with effective dates arriving through this summer.

Don Trone explains how policy, permission, proficiency, platform and protocol keep the industry from reaching artificial intelligence’s full potential

Thrivent shows split in retirement confidence among savers hurdling through market inflation and advancements in AI technology

Candidly releas es Trump Accounts guidance; OakPath launches per sonal coach ‘ Aggi ’ for retirees; and Vestmark o pens AI r esearch c enter.

Everyone in the retirement plan business loves talking about technology. Recordkeepers spend millions promoting participant websites, mobile apps, artificial intelligence tools, and payroll

Third-party administrators (“TPAs”), brokers, and benefits technology vendors are increasingly offering artificial intelligence (“AI”) tools that can generate Summary Plan Descriptions (“SPDs”), plan