Four institutions using four incompatible methods published inside four weeks and pointed the same way: AI exposure rises with national wealth and lands on the young first. The Bank of Korea traced 94% of a four-year fall in Korean youth employment to AI-exposed sectors using pension records rather than a survey. The New York Fed's own firms report almost no AI layoffs. Britain has deferred the survey that could settle it to 2027.
Four institutions, four methods, one direction: AI exposure tracks national wealth and falls on the young first, and the strongest evidence yet comes from pension records, not a survey.
The Bank of Korea counted 285,000 young workers out of insured employment in four years, and found 94% of the loss sitting in sectors it classifies as highly exposed to AI.
Bank of Korea data, published on 18 August, traced 268,000 of a 285,000 youth-job fall since 2022 to AI-exposed sectors. It used pension records, not a survey. Workers in their fifties in the same industries gained 230,000 jobs over the same four years.
The split points to an entry-level squeeze, not a shrinking economy: young workers lose ground in AI-touched sectors while older workers in the same industries hold theirs.
The BIS, the World Bank and the ILO published inside four weeks using methods that share nothing, and all three put AI exposure highest where national income is highest.
United StatesThe Bank for International Settlements (BIS), World Bank and International Labour Organization (ILO) each published AI-exposure findings within four weeks in July and August. The BIS found unemployment rose 0.75 percentage points on average in AI-ready countries between 2023 and 2025. The World Bank put 14.2% of rich-country jobs at risk.
Three institutions using different data reached compatible conclusions: AI's labour effect so far tracks a country's wealth, not its industry mix.
Expansion reported that 2.93 million Mexican workers sit in the highest AI-exposure occupations, and that informality among them runs at 16.5% against 66% at the other end of the scale.
Mexico, como vamos? matched Mexico's labour survey against an AI-exposure index and found 2.93 million workers, 4.9% of employment, in the most exposed jobs, Expansion reported on 11 August. Those roles are mostly formal: informality falls from 66% to 16.5% as exposure rises.
The pattern runs backwards from expectations elsewhere: in Mexico, AI risk concentrates in protected formal jobs, not the country's much larger informal workforce.
Service-firm AI use jumped from 25% to 40% in a year, yet the same firms told the New York Fed they are barely firing anyone. What they are doing is not hiring graduates.
The New York Fed found AI use among US service firms rose from 25% to 40% in a year covering August 2024 to August 2025. Layoffs stayed rare. Hiring suppression concentrated on college-educated applicants instead.
Challenger, Gray & Christmas separately found AI cited in 31% of June's job cuts, rising to 33% in July. The redundancy data tells a sharper story than the survey does.
The ONS pushed its Transformed Labour Force Survey beyond November 2027 and admitted that occupational coding still does not work, which is the coding any AI breakdown of British jobs would need.
The Office for National Statistics said on 11 August its Transformed Labour Force Survey will not launch before November 2027. It also disclosed a May 2026 telephone-collection failure that hit its hours-worked data. Occupational coding, it admitted, remains problematic.
Britain will have no official AI-attribution layer in its jobs statistics for at least another year, leaving the debate to run on private estimates instead.
Challenger, Gray & Christmas counted 33,429 announced US job cuts in July, the lowest monthly total in two years, with artificial intelligence named in 10,970 of them.
Microsoft and Meta shrank headcount in the second quarter while Amazon and Alphabet grew, filings published in late July showed. Challenger, Gray & Christmas separately put AI at the top stated reason for US job cuts for a fifth month running. Its share reached 33% of July's total against 24% year to date.
The overall cut count fell 27% to 33,429: AI's share is rising inside a shrinking pool of layoffs, not growing in absolute terms.
Tencent's quarterly capital spending rose 176% and its free cash flow turned negative by RMB13.8 billion. Baidu's AI cloud grew 50% while its legacy business fell 23%.
Tencent and Baidu's second-quarter results, reported in mid-August, moved in opposite directions on AI capital spending and cash flow. Tencent's capital expenditure reached RMB52.8 billion for the quarter, while Baidu highlighted AI Cloud Infrastructure revenue of RMB7.3 billion.
China's two largest internet groups are now choosing different points on the same capex-versus-cash-flow trade-off already visible among the big US hyperscalers this quarter.
Samsung's chip division posted KRW89.2 trillion of operating profit on AI server and memory demand, while SoftBank Corp. lifted enterprise revenue 11% on cloud and AI work.
Samsung Electronics and SoftBank Corp. both reported AI-driven revenue growth in results published in early August. Samsung's Device Solutions division, which makes memory chips for AI data centres, posted KRW127.5 trillion in revenue for the period.
AI infrastructure suppliers are capturing real revenue growth this quarter even where some of the hyperscalers buying their hardware and services show cash-flow strain.
Siemens will spend more than $200 million on two American factories making electrical equipment for data centres, adding more than 1,500 jobs after taking around EUR6 billion of data-centre orders.
Siemens pledged more than $200 million on 7 August to build two US factories making electrical equipment for AI data centres, adding over 1,500 jobs. The company cited EUR6 billion in data-centre orders over the past nine months.
It is a rare case in AI's labour story of new spending creating manufacturing jobs directly, rather than threatening existing office work.
Italy's public-sector unions and the state bargaining agency signed a national contract on 6 August whose Article 14 bars any exclusively automated decision affecting the employment relationship.
Italy's public-sector bargaining agency and public-sector unions signed a national collective contract on 6 August, adding a dedicated Title on artificial intelligence. The Title sets rules for AI use across public workplaces. It follows a summer of union mobilisation against AI-driven job cuts in Italy's call-centre sector.
A national, bargained AI protection now exists for Italy's public sector at a scale no single-employer deal in recent months has matched.
Justice Suraj Govindaraj extended an interim welfare-fee deposit arrangement to Uber India on 28 July, so platforms now pay the levy into court while the constitutional question waits.
The Karnataka High Court extended an interim arrangement on 28 July, letting Uber India keep making welfare-fee deposits under a state gig-worker law. The underlying legal challenge to the law itself remains unresolved.
Uber's obligation stays in place for now, in Bengaluru, without a ruling on whether the state law can require it at all.
Concentrix told Spanish worker representatives it had lost a further client service, which could take its second redundancy round at A Coruna to 160 posts.
Concentrix may double job losses at its A Coruna site in Spain to 160, reported on 6 August, after losing a further client contract. The first round, 80 jobs agreed in July, was described by unions as Europe's first completed AI-attributed collective redundancy.
The AI framing for this second round again comes from the works committee, not from Concentrix management itself.
Vietnam committed to giving basic AI skills to 10 million workers by 2030. A week later Turkey put 25 billion lira behind a plan promising training for 100,000 workers in two years.
Vietnam and Turkey each launched state AI workforce-training programmes within a week of each other in August. Vietnam aims to give 10 million workers basic AI skills by 2030; Turkey's plan targets 100,000 workers within 24 months, backed by 25 billion lira.
Both governments are training ahead of measured disruption, a rarer sequencing than the job-loss response running through most of this update.
Kenya's draft national AI policy commits the government to a statutory Fair Pay Reference Framework for the data annotators and content moderators who train and police the models.
Kenya's technology ministry proposed, on 4 August, a statutory Fair Pay Reference Framework setting minimum pay for data-annotation and content-moderation workers. Kenyan annotators currently earn $1.46 to $3.74 an hour, against $21 to $27 for comparable US work.
The proposal targets the low-paid workers who build AI systems from underneath, not the ones AI itself may replace.
Cognizant shed 900 staff between two filings and Kyndryl booked a $152 million rebalancing charge with no headcount attached. Neither event produces an announcement anyone counts.
Cognizant and Kyndryl both reduced headcount in results published in early August without announcing a formal redundancy programme. Kyndryl took a $152 million workforce-rebalancing charge, projecting $400 million to $500 million in annualised savings.
Quiet, attrition-driven headcount reduction like this stays invisible to layoff trackers that count only formal, announced cuts.
Direction: up. The specific mechanism that would tip it further is a second national statistical office following Korea's lead and publishing administrative-record exposure data rather than survey data; the ONS's decision to defer its Transformed Labour Force Survey to November 2027 removes the UK as a candidate for at least that period. The countervailing mechanism is the New York Fed's next service-firm survey: if adoption keeps rising past 40% with AI-attributed layoffs still rare, that undercuts a displacement reading and strengthens the hiring-suppression one instead.
