The Philippines' AI Risk Sits in Five Job Titles, Not the Economy

Nearly a million Filipinos share one job code: general and keyboard clerk. All 985,200 of them sit in the single highest generative-AI exposure band the ILO measures. Add two neighbouring clerical codes and you have roughly 78% of the entire country's highest-risk employment inside three lines of an occupational classification.

That is a very different sentence from "AI will take Filipino jobs." It is narrower, more specific, and considerably more useful if you happen to be one of those workers.

The occasion for this is an opinion piece published 17 August 2026 by Khalid Hassan, Director of the ILO Country Office for the Philippines. He asks whether AI will take Filipino jobs and answers that it depends on what the country does now. Fair enough. But an op-ed is a summary of evidence, not the evidence. The evidence is a February 2026 ILO policy brief by Phu Huynh built on Philippine Statistics Authority Labour Force Survey microdata from Q1 2024, and it is far more precise than the column that cites it.

The number that travels, and the number that matters

[Fact] 27.7% of Philippine employment — 12.7 million jobs — has some potential exposure to generative AI. That is the highest rate among ASEAN countries with comparable harmonized data; Indonesia, Thailand and Viet Nam all sit at 21-22%.

[Fact] But only 3.6% of jobs — about 1.7 million — fall into Gradient 4, the band where most tasks are highly automatable and displacement risk is genuinely elevated. The other 11.1 million exposed jobs are in gradients where the technology touches some tasks and leaves the rest alone.

Here is where it gets interesting. That 1.7 million is not scattered across the economy. The ILO lists it: general and keyboard clerks (985,200), sales workers (271,700), numerical and material recording clerks (237,100), other clerical support workers (101,400), and business and administration professionals and associate professionals (75,600). [Estimate] Five categories, summing to about 98% of the national Gradient-4 total. The three clerical codes alone carry roughly 78% of it.

If you are a Filipino data entry keyer, bookkeeping clerk or general office clerk, the national average was never about you. You are the national average.

Why concentration changes the argument

[Fact] More than one-third — 35% — of Philippine clerical support workers are employed in the administrative and support services industry, which is where BPO and contact centres live. Another 9.7% work in finance and insurance, where nearly nine in ten jobs carry some exposure and 22.7% sit in Gradient 4.

The Philippine IT-BPM sector employed 1.8 million workers in 2024, of whom 1.6 million were in contact centres and business process services. So the country's highest-risk occupational population and its flagship export industry are approximately the same size, staffed from the same labour pool, concentrated in the same places — [Fact] 42.4% of jobs in the National Capital Region are exposed, about 2.6 million of them, with Central Luzon and Calabarzon both above 30%.

Most countries can absorb clerical automation because clerks are spread thinly across many sectors. The Philippines built an export industry out of them.

The comfort that does not transfer

Two weeks ago we covered the ILO's ASEAN-wide brief, which found a reassuring paradox: the most exposed workers in Southeast Asia were also the least likely to actually be using AI, with clerical adoption at just 21.1%. Exposure was theoretical because nobody had switched the tools on.

That reassurance does not survive contact with Philippine data. [Fact] A survey by the IT and Business Process Association of the Philippines found that roughly two-thirds of member firms have already incorporated AI tools into operations, while a separate analysis found only 14.9% of Philippine firms overall use them. [Estimate] That is a 4.5× adoption gap running in exactly the wrong direction — the employers of the most exposed workers are the fastest adopters in the economy.

We should say plainly that this revises our own earlier framing. The regional average was true and the regional average was not the whole story.

A promotion with a bill attached

[Fact] On 1 July 2026 the Philippines became an upper-middle-income country, on 2025 Atlas GNI per capita of about US$4,850. The World Bank called it, correctly, an achievement — 11.7 million jobs created over fifteen years and poverty down to 15.5% in 2023.

Six weeks later, on 11 August, the ILO published Global Employment Trends for Youth 2026. Its Table 2.2 reports the share of youth employment in the highest AI-exposure gradients by income group: [Fact] 3.1% for lower-middle-income countries, 7.6% for upper-middle-income countries. The Philippines just moved from the first group to the second.

[Claim] Nobody has put those two August documents side by side, and the connection deserves a caveat: an income-group average is a statement about occupational structure across dozens of countries, not a forecast for one. Crossing a GNI threshold does not reclassify anyone's job overnight. But the direction is not accidental. Countries get richer by moving labour into services, administration and knowledge work — which is precisely the labour that generative AI reaches.

The figure in the op-ed that isn't Philippine

The column cites 6.1% of jobs held by people aged 15 to 29 as highly exposed to AI-related change. We checked it against the source table. [Fact] 6.1% is the World row of Table 2.2 — a global figure, not a Philippine one.

Three different numbers are circulating and they measure different things: the global youth figure (6.1%), the South-Eastern Asia and Pacific subregional figure (8.1%, above the world average), and the Philippine-specific number from the country brief — [Fact] 4.2% of jobs held by 15-24 year-olds, about 217,200 jobs, in Gradient 4. That last one is above the 3.6% national rate. Young Filipinos are over-represented in the highest-risk band even though their overall exposure (27.2%) is marginally below adults' (27.8%).

None of this makes the op-ed wrong. It makes it a column.

Education stops protecting you here

The pattern that should unsettle people who did everything right: [Fact] Filipino workers with tertiary education face 45.4% exposure, with 10.4% — 1.2 million people — in Gradient 4. Workers with only basic education face 21.7% exposure and 1.0% in Gradient 4. [Estimate] The high-risk rate is more than ten times higher for the more educated group.

[Fact] Occupations held by women carry 40.3% exposure against 19.3% for men. Among youth, 7.5% of jobs held by young women are Gradient 4 versus 2.3% for young men.

The most exposed worker in the Philippines is a university-educated young woman doing administrative or customer service work in Metro Manila. That is not the person the phrase "at-risk worker" usually conjures.

What actually happens next

Displacement is the loud scenario. The quiet one is worse for anyone starting out. GET for Youth 2026 spells out that a firm facing automation gains can fire, retrain, or simply freeze hiring — and the third option is "job loss neutral but affecting job creation and potentially blocking job entry for young graduates." It shows up in nobody's unemployment statistics.

Evidence is accumulating. Brynjolfsson and colleagues found a 16% relative employment decline among US workers aged 22-25 in highly exposed fields. A UK study found 4.5% workforce declines at high-exposure firms, concentrated almost entirely in junior roles. Anthropic's 2026 analysis found no unemployment effect but a general slowdown in youth hiring in exposed occupations. PwC calls the mechanism "seniorization": in the most AI-exposed occupations, more than half of the new skills appearing in entry-level postings are ones traditionally associated with experienced workers.

Which is why "skills policy is economic policy" is not a slogan here. It is the operative variable.

If this is your job

The ILO's own framework says exposure converts to outcome through three things: access to the technology, the capability to use it, and employer choices. You control the middle one.

Two data points worth holding onto. [Fact] Across millions of job postings analysed by the ILO, socio-emotional skills accounted for 51.8% of all skills employers listed, cognitive skills 22.9%, and digital and technical skills only 17.7%. And the World Bank's World Development Report 2026, published 4 August, estimates that in low- and middle-income countries generative AI puts 4.5% of jobs at automation risk while meaningfully boosting productivity in 16.2% — [Estimate] a ratio of roughly 3.6 jobs helped for every one at risk, against 1.3 to 1 in high-income countries.

That ratio is the Philippines' actual inheritance, and it is a good one. The country is a genuinely early AI adopter among developing economies, with a workforce already fluent in the service tasks AI is best at assisting. The risk is not that the tools arrive. It is that the tools arrive at the firms while the training does not arrive at the workers — and that entry-level doors close quietly while everyone is watching the headline unemployment rate.

Sources

• ILO, Generative AI and Jobs in the Philippines: Labour Market Exposure and Policy Implications, Phu Huynh, February 2026 — primary source for all Philippine exposure figures; based on PSA Labour Force Survey microdata, Q1 2024
• ILO, Global Employment Trends for Youth 2026: Back to the future, 11 August 2026 — Table 2.2 income-group and subregional youth exposure shares
• ILO, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, Gmyrek et al., ILO Working Paper 140, May 2025 — the gradient methodology all of the above apply
• Khalid Hassan, Will AI take Filipino jobs? The answer depends on what we do now, ILO, 17 August 2026 — entry point and policy framing
• World Bank, World Development Report 2026: The Promise of Artificial Intelligence, 4 August 2026
• World Bank, Philippines Economic Update Midyear 2026 launch remarks, 3 August 2026 — upper-middle-income transition, growth and poverty figures

AI-assisted analysis: this article was researched and drafted with AI assistance. Every statistic was verified against the primary reports linked above, including the correction to the 6.1% youth figure. The 78% clerical concentration ratio, the 4.5× adoption gap, the 10× education gap and the 3.6-to-1 augmentation ratio are our own calculations from the published figures, and we have flagged where an income-group average is being read as a country signal.