Why this work stays with officers
Will AI replace customs and border protection officers in the next decade? The honest answer sits in the task mix, not in the headlines. Primary inspection runs on seconds. An officer reads a travel document, reads a person, asks two or three questions, and decides whether to admit, refer to secondary, or hold. Software can flag a record. It cannot take the oath, make the stop, or stand behind the call in court.
Much of the job is also physical. Searching vehicles, opening containers, handling seized goods and escorting people through secondary inspection all happen in a crowded, unpredictable space. Our robotics read puts a large part of this occupation in the physical bucket, at a tier that would need a dexterous humanoid machine to do it. Nothing on the market works at that level in a port-of-entry lane, and nothing near that level is cheap.
The third piece is authority. Officers exercise lawful discretion: detaining a traveler, seizing property, writing the statement that a prosecutor later reads. Those powers are granted to people, not to systems. Even a perfect screening model would still produce referrals that a person has to work. That is the usual shape of change in law enforcement occupations: the paperwork thins out, the confrontation does not.
What AI does, what it assists, and what it leaves alone
Tasks our data puts in the “AI does it” group cover about 0% of task time here. These are the desk-side pieces: pulling and cross-checking records against watchlists and manifests, and generating routine entry and inspection paperwork. Document readers and biometric matching already run at many crossings, which shrinks the clerical layer of the shift rather than the shift itself.
The assisted group accounts for roughly 52% of task time. Cargo and baggage screening is the clearest example: imaging systems and anomaly detection narrow the pile, and the officer decides what gets opened. Risk-based targeting of travelers and shipments works the same way. The model ranks. The officer still questions the person in front of them and writes the outcome.
What stays with people is about 48% of the work: interviewing travelers at primary and secondary inspection, conducting physical searches of vehicles and property, making arrests and seizures, and testifying about them later. Our overall coverage figure for this job is 26 out of 100 on the question of whether AI can do the work today. You can read how that figure is built on the coverage method page.
What the evidence actually shows
There is no published head-to-head test of an AI system against trained officers on inspection decisions. That is why the parity grade here is D, our marker for untested. We give no parity number when the evidence is that thin, because a guess would read like a measurement. The quality parity method explains the grading scale.
What would settle it is specific: a published trial comparing automated referral decisions with officer referrals on the same traveler and cargo stream, reporting hit rates, false positives, and the time cost of each false positive. Agency deployments of biometric comparison and non-intrusive imaging are real, but deployment is not evidence of parity. Until someone measures both sides on the same lane, the grade stays where it is.
The labor market numbers are firmer. The police and sheriff’s patrol officer group this occupation sits under employed about 670,520 people, with median pay of $76,210 (BLS, 2025), and projected employment change of 3.5% from 2025 to 2035. That is slow growth, not contraction. Hiring pressure in federal inspection roles has come from budgets and staffing policy, not from software replacing posts.
Good to know: cheap software and an officer are not substitutes here, because the software only touches the parts of the shift that happen at a screen.
When the picture could change
Most likely after 2041 (8 in 10 of our scenarios). Two things could pull that earlier. First, automated primary lanes: if biometric entry and exit clears routine travelers end to end, officers shift almost entirely to secondary, and fewer booths are staffed per shift. Second, better cargo anomaly detection, which would cut the number of containers a person has to open before finding anything.
Two things hold it back. The physical share of the work is large and would need robots that can search a vehicle interior, which does not exist at a usable cost. And the cost gap runs the other way from the usual story: the software side is cheap, but it only replaces the screen-based slice, so an officer still has to be paid for the lane, the search and the arrest. Legal authority is the third brake, and the hardest one to engineer around. How we build the window is set out on the replacement-year method page.
How to stay needed at the port of entry
Lean into the tasks that sit in the human group. Interviewing is the first: officers who are good at reading inconsistency in a two-minute conversation produce referrals that hold up. Physical search technique is the second, especially in concealment-heavy work. Testimony and case documentation is the third, because a seizure is only as strong as the statement behind it.
Two skills raise your floor. One is working with targeting systems well enough to know when a flag is weak, which means understanding what the model scored and what it ignored. The other is language ability at your port, which stays scarce and is hard to automate in a loud hall with a nervous traveler.
If you want a nearby move, the closest work sits with police and sheriff’s patrol officers, detectives and criminal investigators, and transportation security screeners. The headline figure for this role is 73 out of 100 (higher is safer), and you can see how every score is built on our methodology page. To put two roles side by side, use the job comparison tool, check the rest of government occupations, or browse the jobs that lean hardest on people.