The DOT Form Policy Shift and What It Actually Changed
In late 2020, the U.S. Department of Transportation finalized its revised Air Carrier Access Act rules, effective January 2021. The rule did two things that reshaped service dog air travel permanently. It limited airline accommodation obligations to trained service dogs only, excluding emotional support animals from the same protections. And it required handlers to submit a DOT-developed attestation form before boarding.
The form itself is straightforward on paper. Handlers attest that their dog is trained to perform a specific task related to their disability, that the dog is not a danger to others, and that the dog will behave appropriately in the cabin. Airlines may additionally require the DOT Service Animal Air Transportation Form for flights of eight hours or more, covering relief behavior.
What the DOT did not specify is how airlines should collect, verify, store or act on that form data. That gap is where the technology problem begins. Five years into the DOT form era, the handler experience remains fragmented and the airline compliance infrastructure is largely paper-based or PDF-based at best. The space is ripe for AI-assisted solutions, but building them correctly requires understanding the regulatory intent first.
Handler-Facing Form Automation: The UX Problem Nobody Solved
The DOT form is a PDF. Airlines accept it via email, printed at the gate or uploaded through carrier-specific portals that were not designed with disability access in mind. For a handler managing a working service dog, luggage and a mobility device simultaneously, the current workflow is a genuine barrier.
Form automation for this context means more than a digital PDF wrapper. It means a system that pre-populates recurring handler information, stores attestation history across carriers, flags upcoming trip requirements with adequate lead time and delivers the completed form in the exact format each airline's compliance team expects. None of the major U.S. carriers have built this. Third-party travel apps have not prioritized it either.
The accessibility dimension matters enormously here. Under the ADA and Section 504 of the Rehabilitation Act, digital government forms must be accessible to screen reader users. The DOT form itself has gone through multiple PDF revisions to address accessibility, but downstream carrier portals frequently strip that accessibility out. An AI-assisted form completion pipeline needs to be WCAG 2.2 AA compliant by design, not as an afterthought.
At ServiceDog.AI, the form automation layer we are developing treats handler data as persistent clinical documentation rather than a one-time transaction. The attestation content is linked to the handler's verified training record, generated from structured data rather than free-text entry, and signed digitally in a format carriers can parse programmatically. That is the architecture the industry needs.
Airline Compliance Systems: Where the Technology Currently Fails
On the airline side, the DOT form created a new compliance obligation without providing technical infrastructure to meet it. Gate agents are expected to review forms, assess whether a dog appears to be a trained service dog and make real-time decisions that carry legal liability under the ACAA. Most agents receive minimal training for this task and have no decision-support tool available.
The result is inconsistency. Some handlers are waved through without form review. Others face extended delays and invasive questioning that exceeds the DOT's own guidance on permissible inquiry. The DOT's two-question rule for service dogs mirrors the ADA's framework: carriers may ask what task the dog is trained to perform and whether the dog is required because of a disability. Nothing more. In practice, compliance varies significantly by carrier, route and individual agent.
Technology can standardize this. A compliance dashboard for gate agents would surface the handler's pre-submitted form, flag whether all required fields are complete, display the permitted inquiry language and log the interaction for audit purposes. That is a relatively simple software problem. The harder problem is behavioral assessment.
Real service dog behavioral standards, as defined by organizations like ADI (Assistance Dogs International) and IAADP (the International Association of Assistance Dog Partners), require that a service dog remain under control, not solicit attention from strangers, not display aggression and not disrupt the environment. A gate agent cannot reliably assess those criteria in a 90-second interaction. Computer vision can help, but it requires careful design to avoid discriminatory application.
Computer Vision and In-Airport Service Dog Verification
The most technically interesting challenge in this space is non-invasive behavioral assessment using computer vision at scale in a real airport environment. Airports are uncontrolled, high-noise environments with variable lighting, crowd density and handler-dog configurations that differ significantly from the controlled laboratory settings used in most animal pose estimation research.
Current state-of-the-art canine pose estimation models, including work published at CVPR and ICCV on quadruped keypoint detection, achieve strong results on benchmark datasets but degrade meaningfully in real-world deployment conditions. Occlusion from luggage, harness equipment and crowd interference are the primary failure modes. Any airport deployment would require a domain-adapted model trained specifically on service dog and airport data.
What computer vision can realistically evaluate in an airport boarding context includes gait steadiness on approach, response to environmental distractors (PA announcements, crowd movement, other animals), handler-following behavior and rest position compliance when the handler is stationary. These are all observable behavioral indicators that trained evaluators use during Public Access Tests. Translating PAT criteria into machine-observable signals is the core research problem.
The ethical constraint is equally important. Any computer vision system deployed at airports must be designed so that a failed behavioral flag triggers additional human review, not automated denial of boarding. The ACAA does not permit airlines to deny transport to a service dog based solely on an algorithmic assessment. Technology here is a decision-support tool for trained staff, not a gatekeeper. Building that architectural constraint into the system from the start is non-negotiable.
Edge inference matters for this use case. A camera-based assessment system running in a jetway or gate area cannot rely on cloud round-trips for real-time decisions. Models need to run on-device at gate hardware or on the airport's local compute infrastructure. That pushes toward quantized, efficient architectures rather than large transformer-based models, even if accuracy trade-offs are involved.
The Canadian CTA Framework: A Regulatory Contrast Worth Studying
Canada's approach to service dog air travel is governed by the Canadian Transportation Agency under the Accessible Transportation for Persons with Disabilities Regulations. The CTA framework takes a different philosophical stance than the DOT's post-2021 rules and the contrast is instructive for technology design.
Where the DOT framework centers on handler attestation, the CTA framework emphasizes carrier obligation to accommodate without requiring advance documentation submission as a general rule. Canadian carriers must accept service dogs that are accompanied by identification documents from an accredited service dog organization, but the CTA also recognizes that not all handlers have organizational documentation, particularly owner-trainers.
The CTA's accessibility regulations also impose specific requirements on the digital interfaces carriers use to communicate with passengers with disabilities. Carriers operating in Canada must ensure their websites and apps are accessible and must not require a person with a disability to use an inaccessible process to request accommodation. That regulatory hook creates a compliance incentive for accessible form technology that the U.S. ACAA framework currently lacks.
For AI system designers, the Canadian contrast reveals an important design principle. A service dog travel technology stack built only to satisfy DOT attestation requirements will be inadequate for cross-border operations. Canadian routes require a system that can handle both attestation-first and documentation-first workflows, recognize accreditation bodies approved under CTA guidance and interface with carrier accessibility compliance teams through accessible channels. Building for the stricter standard first is the architecturally sound choice.
Air Canada and WestJet have each implemented carrier-specific service animal policies that go beyond CTA minimums in some respects. Understanding those carrier-level variations matters for any platform attempting to serve the cross-border handler market. The TheraPetic® Training Plus program, accessible through officialservicedog.com, tracks carrier policy updates as part of its handler preparation curriculum, which gives partnered technology platforms a policy data layer that is difficult to build independently.
Building the ML Pipeline Airlines Actually Need
A production-grade ML pipeline for airline service dog compliance has three distinct layers that must work together. The document processing layer, the behavioral assessment layer and the decision-support layer each have different model architectures, data requirements and latency tolerances.
The document processing layer handles form ingestion, field extraction and completeness verification. This is a well-understood NLP and document AI problem. Models fine-tuned on DOT form variants, carrier supplemental forms and handler-provided training documentation can achieve high accuracy on field extraction. The output is a structured data record, not a raw document, that feeds downstream systems.
The behavioral assessment layer is the novel contribution. As described above, this requires a canine pose estimation model adapted for airport environments, a temporal action recognition component to evaluate behavior sequences rather than single frames, and a confidence scoring system that quantifies uncertainty for the human reviewer. Low-confidence assessments should always escalate to human review. The model should be explicit about what it observed and what it could not observe given environmental conditions.
The decision-support layer is a presentation layer, not an AI layer. It takes outputs from the first two layers and presents them to a gate agent in a format that supports a legally compliant, time-bounded decision. It surfaces the two permitted DOT inquiry questions, the form completeness status and the behavioral assessment summary. It does not make a binary allow-deny recommendation. That decision remains human.
Integration with carrier departure control systems is the deployment challenge. Major carriers run proprietary DCS platforms that do not expose standardized APIs for third-party compliance tools. SDK-based integrations, browser extension overlays or tablet-based companion apps are the realistic entry points for a startup building in this space. Airline technology procurement cycles are long. Building a handler-facing product first, generating handler adoption data and then approaching carriers with demonstrated use cases is a more viable go-to-market path than a carrier-first approach.
Policy, Technology and the Handler Experience
The DOT form era formalized something important: service dogs in aircraft cabins are accommodations granted to handlers with disabilities, not informal permissions extended by airline discretion. That policy clarity is valuable. The technology infrastructure to implement that policy fairly and consistently has not kept pace.
Handler-facing form automation, carrier compliance dashboards, computer vision behavioral assessment and cross-border regulatory compatibility are all solvable problems with current AI and software tooling. What is missing is a development organization that treats the handler's experience as the primary design constraint rather than the carrier's operational convenience.
ServiceDog.AI's work at the intersection of computer vision, clinical documentation and disability rights policy is oriented toward that design priority. Service dogs are working medical equipment. Their handlers have legal rights that predate any airline preference. Technology built in this space must enforce that reality at every system boundary.
Organizations like IAADP, ADI and the verification resources at officialservicedog.com and officialserviceanimal.com represent the practitioner knowledge base that any AI system in this domain needs to be grounded in. The clinical AI layer at therapetic.ai provides the handler health documentation infrastructure. The canine behavior and task performance assessment is the domain-specific contribution that ServiceDog.AI is positioned to develop.
The DOT form problem is, at its core, a data interoperability problem wearing a disability rights costume. Solving it correctly requires technical rigor, regulatory literacy and genuine commitment to the handler community. In 2026, the tools exist. The will to build them correctly is what the field is still developing.