Skip to content

Automated Customer Service Systems



Rule status: Proposed

Agency: DCWP

Comment by date: December 18, 2026

Printable Version of Proposed Rule Text
DCWP-NOH-Proposed-Rules-Regarding-Automated-Customer-Service-Systems.pdf

The Department of Consumer and Worker Protection is proposing to add rules to prohibit certain deceptive or unconscionable trade practices prevalent in the provision of customer service due to the increasing use of automated customer service systems. The Department is also clarifying that no person can avoid liability for failing to comply with local laws and rules by claiming a particular technology caused or assisted in violative acts or omissions.

Send comments by

Public Hearings

Attendees who need reasonable accommodation for a disability such as a sign language translation should contact the agency by calling 1 (212) 436-0183 or emailing [email protected] by November 12, 2026

Date

December 18, 2026
11:00am - 12:00pm EST

Connect Virtually
https://tinyurl.com/5e2m7k2c
Please dial +1 646-893-7101
Phone conference ID: 886 506 734#
Meeting ID: 213 525 864 272 361
Passcode: PC7nd79A

Disability Accommodation
  • Sign Language Interpretation
  • Open Captioning
  • Communication Access Real-Time Translation

Comments close by December 18, 2026

Add a comment

Notes. "Required" indicates a required field. Your email address will not be made public.

Online comments: 21

  • Anonymous RD

    I strongly support this proposal to create common-sense restrictions on AI slop!

    Comment added October 20, 2026 2:37pm
  • Lauren

    I also strongly support this. There is nothing more frustrating than getting the runaround from an automated system and becoming frustrated because you cannot get an answer. I want the option of talking to a human being.

    Comment added October 20, 2026 3:25pm
  • Dario Tonelli

    Start with transparency: it should be illegal for a chatbot to present itself as a person. Not just in customer service, that should be a broadly applicable rule (and it should work both ways, a consumer who sends an agent to inquire should be required to be equally transparent — which means, the companies that offer those agents should be required to make them transparent)
    Direct access to a person: if someone asks to speak to a human, they should be directed to a human. For small companies that do not have the scale to offer customer service, there could be an exception, but again it should be transparent: if someone asks to speak to a human and that option is not available, the response should be “No, human reps are not available” instead of stringing people along.
    That should be limited to small transactions. If a company sells an app for $1.99, it is reasonable to deny customer service from humans. If someone buys a $2,000 appliance, the bar should be different.

    Comment added October 20, 2026 9:12pm
  • Madeline

    I am writing in strong support of this proposal. I have experienced purposely misleading and useless customer services features that are designed to make customers give up on receiving support (most recently Lyft, where following prompts to report an errant charge led me to a satisfaction survey rather than a refund request form). I’m thankful these rules are being proposed and I look forward to a future of customer service that actually provides help, not protects the interests of the company.

    Comment added October 21, 2026 10:44am
  • Gabriela Maciel

    This is long overdue, so thank you! So many times I feel folks get so frustrated with these menus and chat bots that they simply give up. I hope this passes. You’ve got my vote!

    Comment added October 21, 2026 6:45pm
  • Anonymous

    Finally this is getting addressed! I strongly support this proposal. I deal with this constantly and dread having to call doctors, insurance companies, etc. for this exact reason. It shouldn’t be this difficult to get a live person on the phone. I feel especially terrible for the customer service reps who have to deal with irate customers that wouldn’t be so irate if they didn’t have to go through 10 prompts to get a human on the line! Thank you so much for addressing this. I really hope it passes.

    Comment added October 21, 2026 8:35pm
  • Anonymous

    If I had to pick between sitting in 3 hour standstill traffic on the Cross Bronx in 95 degree weather or returning a missed call from Walgreens/CVS Pharmacy, I’d pick the Cross Bronx traffic even if the car didn’t have air conditioning. It should not take 4+ minutes of yelling at an AI phone answering service in order to simply leave a voice message for the pharmacy staff with the additional information they asked for to fill your prescription. If there’s an insurance issue or a dosage issue, the endless phone answering machine loops force patients/consumers to either go in person to communicate that information even if their prescription isn’t ready or to forfeit the prescription entirely because there’s no way to get in touch with the pharmacy staff. I don’t need a direct line to talk to a human being- just a voicemail inbox to leave a message so they can continue filling my prescription when they get to it.

    Comment added October 22, 2026 2:17am
  • Cary Raffle

    We are all frustrated by the difficulty getting help on on customer service calls. Recently, I called into AppleCare with a problem and to my surprise, the AI customer service bot provided better and faster service than I have received from many customer service calls. It quickly answered my questions, understood and spoke perfect English, did not read annoying scripts or thank me over and over for my loyalty and patience, It solved my problem quickly and efficiently. There are very few call center jobs in NYC, so this rule will protect our workers. It will however put us in a position where we may continue to receive the worst customer service in the world. I oppose the bill for these reasons.

    Comment added October 22, 2026 8:35am
  • Sally Jane Gellert

    This is a common-sense proposal to help customers express their needs to business owners. If a chatbot works, then fine; but when it doesn’t work, then there needs to be easy access to a human. As a Bergen County resident who at times deals with New York City businesses, I support this rule-making and would like to see it instituted in other jurisdictions as well, including here in New Jersey.

    Comment added October 22, 2026 7:28pm
  • Merv Cae

    I am a senior citizen of New York City and I have problems navigating the robot system that some businesses use so please please implement the rule that businesses must provide humans to assist customers over the phone thank you MC

    Comment added October 23, 2026 9:55am
  • Anonymous

    I am so grateful for this proposal. For all of us, but especially for those who deal with serious medical or financial issues, poor customer service threatens our energy, resources, and basic rights.

    My ideal would be:
    *all chatbot use immediately and clearly named at the start of an interaction
    *instructions shared immediately for how to speak with a qualified person
    *some easy, universal system to monitor, record, and address any violations of the above
    *create a gold star registry of strong customer service in NYC to incentivize companies to value it and help citizens reward it

    Please hold companies accountable for using AI slop that leads to real negative impact. Thank you for not succumbing to cynical narratives of AI as the future by default, and taking action to lead with our values first.

    Comment added October 24, 2026 9:46am
  • Anonymous

    Companies like SevenRooms (owned by DoorDash) , Slang AI, and Toast are now selling “Voice AI” agents to restaurants for answering the phone. There should be an incentive for restaurants to use humans instead of these ai agents. SPIN flatiron is a good example of a SevenRooms voice agent – 48 E 23rd St, New York, NY 10010

    Comment added October 26, 2026 11:22am
  • Anonymous

    This proposal is a step forward in consumer protections and should be supported. Companies have been pivoting towards AI and automated customer service chatbots which very often fail to provide any meaningful resolution to the problem. Without human intervention, the customer is often left without any real recourse and the city should intervene to protect consumers.

    Comment added October 26, 2026 2:24pm
  • stacie johnson

    Yes, this happened to my when I tried to call NYCERS, the NYC employee retirement system. I hope all city agencies and it’s affiliates will be subject to improved customer service experience by prioritizing human representatives.

    Comment added October 29, 2026 10:46am
  • IR

    Will this apply to NYC and NYS government offices as well?

    Comment added October 29, 2026 6:13pm
  • Tim Armacost

    I wholeheartedly support this initiative to rein in abuse of AI systems. AI cannot take into account factors that fall outside of the rules as defined for them.

    My insurance company makes me give the chatbot my address, date of birth etc, in order to enter the system, and then the human representative asks me for all of it again. Every time. The companies may think they are saving themselves time, but they are wasting mine. Thank you for attempting to bring some balance back to this system.

    Comment added November 1, 2026 1:17pm
  • Marla Tepper

    I strongly support the Department’s proposed rule and appreciate the Department’s efforts to address this important issue. I hope the Department will consider the suggestions in my attached comments, which are intended to strengthen the proposed rule and further its objectives. Thank you!

    Comment attachment
    Comments-Re-DCWP-Proposed-Rule-on-Automated-Customer-Service-Systems-MTepper-11.5.26.pdf
    Comment added November 5, 2026 11:51pm
  • Sam NYC

    Everyone has experience with customer support, the good and the bad. The reality is that human-led customer support has been degrading for decades. Today, even when we reach humans, I question their knowledge and efficiency. This is due to heavy off-shoring and poor training in an effort to save money. Faster access to humans is not the answer if those humans aren’t good at what they do.

    At the same time, the AI customer support landscape is evolving quickly. I have had amazing experiences getting personalized support from AI agents used by some of my favorite NY-based companies. These chat experiences are specific to me, knowledgable, and solve my problems far faster and more professionally than a human could.

    We will always need and want humans in the loop, but the answer is not to incentivize even more emphasis in low-quality, slow, off-shore human call centers, but to empower NY’s businesses to meet customers where they are with high-quality AI capabilities that improve customer experiences.

    I agree to having the right guardrails and transparency to help businesses improve their agentic work, while still leveraging this crucial, forward-looking, and customer-centric innovation.

    Comment added November 10, 2026 3:30pm
  • Vasanth Rajendran

    Comment on Proposed Rules: Automated Customer Service Systems

    To: New York City Department of Consumer and Worker Protection
    Re: Proposed amendments to Title 6 of the Rules of the City of New York (6 RCNY § 5-02; Subchapter U, §§ 5-310 to 5-313; § 6-91), published in the City Record on October 20, 2026
    From: Vasanth Rajendran
    Date: November 10, 2026

    Introduction

    I am a Software Development Manager at Amazon with 14+ years in software engineering and engineering leadership. I chair the IEEE P1968 Working Group, which is developing a recommended practice for the governance of autonomous AI agent systems, and I serve as a Privacy Bar Raiser at Amazon, an internal reviewer with sign-off authority on privacy design. I am an IEEE Senior Member and a Fellow of the Institution of Electronics and Telecommunication Engineers (IETE). I authored the SAFE-AGENT-L framework paper at the AAAI-26 AI and Law Bridge and the companion open-source library, and an ICAITech paper on AI-powered portable screening for diabetic retinopathy. I am an invited IEEE speaker and a reviewer for NeurIPS, ICLR, and IEEE AI conferences, and I have commented on AI governance to NIST, the Colorado Attorney General (ADMT Act and Chatbot Safety Act), TRAI, and BEREC.

    My work centers on a practical question this rule answers directly: when an automated system speaks for a business, who is accountable for what it says, what is it permitted to decide, and how does a person get out when it fails?

    What the rule gets right

    The proposed rule turns four sound governance principles into enforceable obligations. First, identity is not authority: § 5-312(a) requires the merchant to disclose, clearly and conspicuously at the start of a covered interaction and in the interaction’s language, that the consumer is dealing with an automated system, not a human. Second, authority must be bounded: § 5-312(b) names the doom loop, a merchant that represents service as available and then fails to provide it reasonably and timely, including repeated, unreasonable routing to menus that cannot help, or unresponsive, irrelevant, or erroneous answers. The Department cites a Twilio survey finding that 40% of consumers say AI repeats itself or gets stuck in loops. Third, a human offramp is a control, not a courtesy: the rebuttable presumption in § 5-312(c), paired with the single-step transfer mechanism in § 5-310, gives consumers a reliable exit and merchants a clear compliance target. Fourth, controls need an audit trail: § 5-313 requires machine-readable records and annual aggregates, including a 72-hour repeat-contact count that detects loops at scale. Above all, § 5-02 removes the technology-blame defense: reliance on technology is not a defense to a violation, so a business remains responsible for conduct that violates applicable requirements, just as for its human staff’s statements. This is consistent with the Air Canada chatbot matter the Department cites.

    Recommendations

    1. Keep the rebuttable presumption and treat the single-step transfer as a bounded-authority control (§§ 5-310, 5-312(c)).

    An automated system that cannot hand a consumer to a human is, in effect, exercising authority to close the matter. No one granted it that authority. A merchant may authorize a system to manage a bounded interaction, but not to hold a consumer in the loop once the consumer asks for a human. I urge the Department to retain the presumption as proposed and reject any substitution of disclosure for transfer.

    The transfer mechanism should remain available throughout the interaction, and repeated requests for a human that are ignored or intercepted by further automated questioning should count as evidence of noncompliance. A transfer offered only after the consumer completes a set number of automated attempts, or only after the system “tries to help first,” is not a single step. The routing-menu exception in § 5-310 should stay narrow: I ask the Department to clarify that a menu that only selects the right department qualifies, but a prompt asking the consumer to describe the issue, or any substantive troubleshooting before transfer, does not.

    One clarification: § 5-312(c) is a rebuttable presumption operating through § 5-312(b), not an unconditional violation, and it permits one of seven alternative means when a human is unavailable or the queue is unreasonably long. Those alternatives must still provide either the hours during which a human representative is available or a stated response timeframe, as the rule requires for each alternative. The rule does not guarantee an immediate human conversation in every case.

    2. Require that a transfer carry the interaction’s context (§§ 5-312(c), 5-313).

    When a consumer activates the single-step transfer, the human representative should receive a short record of the automated interaction: the consumer’s stated issue, what the system attempted, and any information it provided. I describe this as a delegation receipt for the handoff. Without it, the consumer must re-explain from zero, often after re-verifying identity. A transfer that drops context is a second loop.

    I recommend that the Department require the merchant to make this record available to the representative at the time of transfer, and add a § 5-313 field recording whether the handoff record was delivered to the representative at the time of transfer. The record should cover the stated issue, information already collected, actions attempted, and unresolved questions or failures. It need not expose proprietary model reasoning, and it must not carry passwords, authentication secrets, or unnecessary sensitive information. Consistent with data minimization, it should carry only what the representative needs.

    3. Log failed transfers and failed overrides as distinct fields (§ 5-313).

    The proposed per-interaction fields record whether a transfer was offered, activated, and whether the consumer was transferred, queued, or given an alternative. They do not capture what happens when the override fails, and a failed override is the most important event to record, because it shows the safety mechanism did not work when the consumer needed it.

    Queuing satisfies the transfer mechanism as § 5-310 defines it, since the mechanism initiates transfer or queuing. The gap is in the records: logs should distinguish a consumer who was queued and then connected from one who was queued and then abandoned, so a queue is not counted as a successful transfer when no human ever answered.

    I recommend fields for: (a) a transfer requested but not initiated (ignored or blocked); (b) a transfer activated but never connected (dropped, timed out, disconnected); (c) a queue entered and then abandoned, with wait time; (d) repeated transfer requests in one interaction; (e) an alternative means under § 5-312(c) unavailable during stated hours or a promised response not provided within its stated timeframe, as applicable; and (f) the failure reason where available from system logs. The annual aggregates should count each. These fields should record observable events: abandonment is evidence warranting examination, not automatic proof of a violation.

    4. Use the 72-hour repeat-contact metric as a public loop detector (§ 5-313).

    The 72-hour repeat-contact count is the right signal: a loop shows up as a consumer who comes back, and comes back again. I recommend three refinements. First, define how a repeat contact is identified (same account, phone, or email within 72 hours) so merchants count consistently. Second, define “resolved” so an abandoned interaction does not count as resolved without a transfer; otherwise the metric rewards systems that wear consumers down. The aggregates already count interactions resolved without a single-step transfer, so the gap is the missing definition, not a missing field. Third, require merchants above a size threshold to submit their annual aggregates, and ask the Department to publish the repeat-contact, resolved-without-transfer, and failed-transfer rates. The rule requires merchants to keep records and produce them to the Department upon request, not submit them routinely, so a submission step is needed first. Any size threshold for submission should be set on evidence and should not relieve smaller merchants of the underlying recordkeeping duty. Published figures should be aggregated so they do not expose individual consumers. These measurements would not independently establish deceptive conduct; they would provide indicators for monitoring, investigation, and comparison. Without public visibility, the aggregates remain internal compliance records.

    5. Extend the disclosure to cover scope, not only identity (§ 5-312(a)).

    Knowing a system is automated is necessary but not sufficient; consumers also need to know what it can do. I recommend that the opening disclosure state plainly, in the interaction’s language, what the system can and cannot resolve, along with the transfer mechanism. For example: “I’m an automated assistant. I can track orders and start returns. I can’t change billing or approve refunds. For anything else, say ‘agent’ at any time.”

    A chatbot may explain a refund policy but lack authorization to approve a refund, or retrieve billing information but lack the ability to correct a charge. Such disclosures should be concise and accessible, not lengthy technical notices. I recommend extending the principle underlying § 5-311, which bars misrepresenting which channels offer customer service, to misrepresentations about what an automated system can do. I also recommend that when a system recognizes a request outside its stated scope, it offer the transfer rather than attempt an answer. That complements, not replaces, the consumer’s ability to request a transfer at any time (Recommendation 1).

    6. Add a correction duty for erroneous information (§§ 5-312(b), 5-313).

    Section 5-312(b) correctly treats repeatedly and unreasonably providing erroneous information as a failure to provide service. The rule should also say what happens next, as a proposed addition rather than an interpretation of the existing prohibition. Where an automated system gives a consumer wrong information about prices, fees, billing, refunds, eligibility, or account status, the merchant should correct it in writing within a reasonable period once it learns of the error, whether from the consumer or from its own review, and log the correction under § 5-313. I ask the Department to set that period in the final rule.

    The burden should sit with the merchant, which holds the retained records, not with the consumer. The merchant should investigate reasonably supported error reports using its own records, without demanding that the consumer prove why the system erred. Where a consumer needs to dispute a material error, I recommend access to the relevant retained interaction information, subject to identity verification, redaction, and privacy requirements. This is a proposed addition: § 5-313 requires interaction logs and compliance records, not necessarily complete transcripts. Read with § 5-02, a merchant should not deny a remedy for reasonable reliance on the ground that the statement came from its automated system, though § 5-02 removes a defense and does not itself create a universal right to reimbursement. The duty should be proportionate, distinguishing consequential inaccuracies from minor wording mistakes, and records should capture whether the error is isolated or recurring.

    7. Read § 5-02 together with vendor accountability (§§ 5-02, 5-313).

    I strongly endorse § 5-02. I ask the Department to state in the final Statement of Basis and Purpose that a merchant’s contracts with its technology vendors do not change its responsibility to consumers: a merchant cannot disclaim a violation by pointing to its chatbot provider.

    Section 5-313 already requires records related to system acquisition, development or installation and the periods during which specific models were used. I recommend building on that obligation: link each covered interaction to a deployment or configuration identifier, and retain, for each version, the vendor’s documentation of guardrails, escalation rules, and known failure modes. A version identifier alone may not identify deployed behavior when prompts, tools, routing, or policy settings change independently, so it should capture the configuration that produced the interaction. The merchant should hold contractual and technical access to vendor-held compliance evidence when reasonably required for an investigation. The obligation should target compliance-relevant records, not unrestricted disclosure of a vendor’s technical internals or proprietary algorithms.

    Conclusion

    With this rule, New York City would, as the Department describes it, be the first municipality in the nation to give consumers an enforceable path to a human when automated customer service fails. The rule’s structure is sound: disclosure, a bounded system, a single-step offramp, an audit trail, and clear deployer responsibility. My recommendations aim to keep those protections from being hollowed out in practice.

    Thank you for the opportunity to comment. I would be glad to provide further detail on any of these recommendations.

    Respectfully submitted,

    Vasanth Rajendran
    Software Development Manager at Amazon
    Chair, IEEE P1968 Working Group on Autonomous AI Agent Systems Governance
    [email protected]

    Comment added November 11, 2026 1:54am
  • Anonymous

    I strongly support this Proposed Rule!
    Automated “customer service” menus rarely address the reason for my call and send me on a time-wasting goose chase trying to get the help I need. Quite often my issue or question needs to be explained to a human and doesn’t fit neatly into a pre-determined category. Often there is no menu option given to talk to a human (even if you follow the menus to their conclusion) and yet if I finally press “0” or say “representative” or “talk to a human” I’m transferred to someone (even though that option wasn’t offered)–this strikes me as dishonest and manipulative when the ability to talk to a human exists, but the company won’t tell you how to access them. I sometimes think the hassle and waste of time is done deliberately so the consumer will just give up and the company won’t have to grant a return or refund or resolve a billing issue. I also strongly agree that an AI agent should identify itself as such.
    Thank you for considering my position in favor of this proposed rule.

    Comment added November 12, 2026 7:53pm
  • Anonymous

    Overuse of automated systems has become customer avoidance, not customer service. Once a person says “human being” or “representative” they should be able to reach a human because the AI bot is not able to be of assistance, which I often find with complex matters. Most humans can figure out when they need a human for an issue, and should not be subjected to phone torture.

    Comment added November 13, 2026 2:54pm
[shared_counts]