No expectant moms at the border: Trump's birthright Plan B - Axios

The Supreme Court Ruling and the Birthright Citizenship Backlash

The 6-3 decision in June 2025 was unambiguous: "All persons born or naturalized in the United States,? And subject to the jurisdiction thereof, are citizens? " Justice Jackson's concurrence rebuked Thomas's dissent by citing the original understanding of the Citizenship Clause. Yet the administrative state moves faster than the judiciary. Within 48 hours, the White House issued an executive order directing DHS to "maximize the use of existing statutory authority to discourage birth tourism. " The key word is "existing. " No new law, just software,

Axios's reporting - No expectant moms at the border: Trump's birthright Plan B - detailed how CBP will now flag any female visa applicant whose travel history includes a "maternity tourism" cluster. That cluster was trained on datasets from private insurance claims, hospital registrations, and even social media geotags. The system doesn't need a confession; it uses a Bayesian classifier to predict pregnancy probability based on arrival date and departure dates. As one former DHS data scientist told me, "It's like a fake news detector. But for baby bumps. "

What Is Trump's "Plan B" for Birthright Citizenship? A Technical Decomposition

Plan B rests on three technological pillars: predictive risk scoring - biometric tracking. And retrospective penalty enforcement. The first pillar, predictive risk scoring, was built by modifying the DHS Office of Biometric Identity Management (OBIM) algorithms. Traditionally, OBIM verifies identity against known terrorist watchlists. Now its neural network ingests features like "country of origin fertility rate," "age," "marital status," and "number of previous U. S medical visits. " The output is a "maternity risk score" from 1 to 100.

The second pillar - biometric tracking - is where the engineering gets hairy. Under the Birthright Citizenship Law (8 U. S, and c§ 1401), a child born in the US acquires citizenship automatically. Trump's team argues that if the mother entered with a "material misrepresentation" (i. And e, lying about intent to give birth), then the child's citizenship can be retroactively challenged. To prove intent, CBP now cross-references hospital birth records with the mother's entry declaration using a graph database (Neo4j) that links every woman who gave birth within 9 months of entering on a B-2 visitor visa. The database update schedule? Real-time, via an API from the National Vital Statistics System. This is a massive engineering achievement - and a privacy nightmare.

The Axios article quoted a senior DHS official saying, "We can now identify 92% of birth tourism cases within 3 days of birth. " That number comes from a precision-recall curve optimized on a test set of 40,000 historical records. But as any ML engineer knows, a production model with 92% precision at 60% recall means 40% of real birth tourists slip through - and an unknown number of legitimate visitors get flagged as false positives. Those false positives become the "No expectant moms at the border" reality: women with valid visas turned away at airports because the algorithm predicted pregnancy.

AI-Powered Surveillance: Drones, Sensors. And Predictive Analytics Go Mainstream

Plan B doesn't stop at visa offices. CBP has expanded its use of Aerostats and MQ-9B SkyGuardian drones along the southern border, now equipped with multispectral sensors that can detect body heat changes indicative of late-term pregnancy. According to a leaked procurement document from February 2025 (since verified by The Guardian), the "Maternity Detection Payload" uses a YOLOv5 object detection model trained on thermal imagery of pregnant mannequins and real volunteers. The system alerts agents within 200 meters.

This is the same class of technology used in precision agriculture to count crops. But repurposed for human biology. The engineering challenge is immense: thermal signatures vary by gestational stage - ambient temperature, and clothing. The model's false-positive rate at the border during summer months (when abdominal temperatures rise) is currently 15%, meaning one in seven alerts is a false alarm. That's acceptable to CBP. But it means 15% of those "no expectant moms at the border" stops are women who are simply overweight or wearing a heavy sweater. Privacy advocates have already filed an ACLU suit citing Fourth Amendment violations.

The Birth Tourism Industry and Digital Detection: A Cat-and-Mouse Game

The "birth tourism" industry - comprising agencies in China, Nigeria. And Russia that charge $40,000+ for "U. S citizenship packages" - has responded with its own technical countermeasures. Online forums now share tips to evade the algorithm: "Avoid searches for 'obstetrician' on Google before your trip," "Don't like any baby product ads on Facebook," "Enter the US 10 weeks before your due date, not 8. " These are adversarial examples directed at the feature space of the DHS classifier.

From a software engineering perspective, this is a classic arms race. The DHS model must now incorporate "adversarial training" - adding synthetic examples of evasive behavior to the training set. But each update requires re-training on a dataset that includes millions of visa applications. Which takes weeks on a GPU cluster (likely AWS p4d instances). The cycle time means that well-resourced birth tourism networks can adapt faster than government procurement allows. The Axios piece quotes an anonymous DHS engineer saying, "We're always three steps behind because we can't roll out a model update in less than a month. " This isn't just a policy story; it's a DevOps story.

Algorithmic Bias and Due Process Concerns in Government AI

Any engineer who has built ML systems knows the bias debt. The DHS pregnancy classifier was trained on a dataset that's 78% Chinese and 15% Nigerian - reflecting the demographics of known birth tourism agencies. But when deployed globally, it disproportionately flags women from all low-income countries, even if they have no intention of giving birth. An internal DHS audit found that women from El Salvador are 4x more likely to be flagged as "high risk" than women from Japan, despite the latter having a higher per-capita birth tourism rate. The model learned that "country of origin fertility rate" is a proxy for race.

The fix sounds simple: de-bias the model by removing demographic features. But the model's accuracy drops from 92% to 67% without those features. And cBP chose accuracy over fairnessFor engineers, this is a reminder that algorithmic systems embedded in law enforcement inherit the worst biases of the humans who built them - and no amount of hyperparameter tuning can fix a flawed policy objective.

Software Engineering Challenges in Building Immigration Systems at Scale

Consider the architectural challenges. The "birth tourism detection" system must ingest data from: (a) the State Department's Consular Consolidated Database (CCD) - a 30-year-old mainframe running COBOL, (b) the CDC's National Vital Statistics System - a modern REST API, (c) hospital admission records from 5,000+ US hospitals - each with different HL7 formats, and (d) social media APIs - often rate-limited. The integration layer uses Apache Kafka for event streaming, with schema registries for each source. The latency requirement: a visa applicant must get a decision within 24 hours of submitting their DS-160 form.

A single misaligned timestamp - say, the hospital records system uses epoch while CCD uses ISO 8601 - could cause a legitimate visa holder to be flagged as a "late arrival" and denied entry. During testing, engineers found a bug where a woman who had a miscarriage was still flagged because the system only looked for "birth records" without checking for "stillbirth" codes. The DHS dev team (contracted through Booz Allen) had to add an extra filter for ICD-10 codes Z37. 1 (single stillbirth). This is the reality of coding for life-or-death decisions.

The Geopolitical Implications of Automated Border Control

Axios's coverage has been widely cited by international media, particularly in China and India. Where the "Plan B" is seen as a digital wall. The Chinese Foreign Ministry announced it would issue travel advisories warning pregnant women that "U. S algorithms may subject you to privacy-invasive screening. " This threatens the $1. 2 billion Chinese birth tourism industry and could strain diplomatic relations. In response, the DHS has begun geofencing visa applications from countries with high retrieval rates, essentially treating the algorithm as part of foreign policy.

The NPR article on the Supreme Court ruling noted that other countries are watching closely. Several nations with birthright citizenship (e, and g, Canada, Brazil) are considering similar algorithmic tools. The difference is that Canada's system requires a court order before accessing hospital records. The US model, under Executive Order, bypasses judicial oversight. For engineers, this is a case study in how software architecture can circumvent constitutional protections.

Ethical Considerations for Developers Building Government Systems

As a contractor, you may be asked to implement a feature that discriminates - even if the code is mathematically correct. The ACM Code of Ethics states: "Software engineers shall act consistently with the public interest. " Building a pregnancy-detection algorithm without a fairness audit violates that principle. Yet the DHS contract stipulates that all model outputs are "law enforcement sensitive" and exempt from public review. If you're a senior engineer considering such a role, ask: Can I live with the consequences of a false positive that separates a mother from her newborn?

The open-source community has already responded. A group of former FAANG engineers launched "Birthright Watch," an independent monitoring tool that scrapes FOIA requests and CBP press releases to track false-positive rates. They use a BERT-based natural language model to parse CBP operational reports and flag discrepancies. Their first analysis, published last week, estimates that 1,200 women have been denied boarding since Plan B launched solely based on algorithmic flags - and fewer than 10% had actual birth-tourism intent. The tool's code is on GitHub; you can contribute.

Frequently Asked Questions

• What is the technical difference between the original Trump birthright order and "Plan B"?
  The original Order tried to reinterpret the 14th Amendment by Executive fiat. Plan B uses ML models to assert that mothers who "intended" to give birth committed visa fraud, thus making the child's citizenship retroactively invalidatable. It's enforcement by software, not by law.
• How does the DHS pregnancy prediction model work?
  It's a gradient-boosted decision tree (XGBoost) trained on features like travel dates, age, nationality, medical history fields in visa applications. And geofenced hospital admission data. Output is a risk score. The model is updated monthly using new birth records.
• Can a flagged visitor appeal before being removed.
  In theory, yesIn practice, the expedited removal process allows CBP to deport within 48 hours without a hearing. The ACLU is challenging this as a violation of procedural due process.
• What open-source tools are available to audit such systems?
  The AI Fairness 360 toolkit from IBM, Aequitas. And the "Birthright Watch" FOIA parser (GitHub: birthright-watch/foia-cbp) can help analyze model outputs. However, DHS hasn't released its model weights or training data.
• How can a developer ethically work on government immigration software?
  Insist on a published fairness audit, demand a kill switch for false-positive victims. And include a public API for independent monitoring. If the client refuses, consider whistleblower channels or refuse the contract.

Conclusion: Code Is Policy, and Policy Has to Be Coded Right

The Supreme Court's ruling upheld the Constitution, but technology is now being used to quietly undermine it. The case of "No expectant moms at the border: Trump's birthright