
The IMF’s AI Deployment Report Puts Productivity Claims on an Evidence Clock
The IMF’s 2026 AI analysis is most useful when read as a warning about adoption, labor, and measurement rather than a forecast of automatic productivity.
The IMF’s AI Deployment Report Puts Productivity Claims on an Evidence Clock
Executives can buy an AI tool in an afternoon. Economies cannot transform that quickly. The IMF’s 2026 annual-report chapter on AI deployment and disruption places the debate where it belongs: between technical possibility and measured economic change. The question is no longer whether a model can complete a task in a demonstration. It is whether organizations redesign work, train people, and record enough evidence to know when output has improved rather than merely accelerated.
The report separates exposure from transformation
The IMF’s “AI: Deployment and Disruption” chapter is the primary source for its framing of adoption, productivity, and labor-market effects: https://www.imf.org/annual-report/2026/in-focus/ai-deployment-and-disruption/. Exposure to AI does not equal displacement, and access to a model does not equal productive use. A role may contain automatable tasks while the organization lacks clean data, workflow authority, or the ability to verify results. That gap is where most real deployment economics will be decided.
The IMF’s “AI: Deployment and Disruption” chapter is the primary source for its framing of adoption, productivity, and labor-market effects: https://www.imf.org/annual-report/2026/in-focus/ai-deployment-and-disruption/. Exposure to AI does not equal displacement, and access to a model does not equal productive use. A role may contain automatable tasks while the organization lacks clean data, workflow authority, or the ability to verify results. That gap is where most real deployment economics will be decided. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
The IMF’s “AI: Deployment and Disruption” chapter is the primary source for its framing of adoption, productivity, and labor-market effects: https://www.imf.org/annual-report/2026/in-focus/ai-deployment-and-disruption/. Exposure to AI does not equal displacement, and access to a model does not equal productive use. A role may contain automatable tasks while the organization lacks clean data, workflow authority, or the ability to verify results. That gap is where most real deployment economics will be decided. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Productivity is a process outcome
A faster draft is not necessarily more output if review expands, rework increases, or downstream systems cannot accept the result. The OECD’s work on AI and productivity makes the complementary point that organizational capabilities mediate technology gains: https://oecd.ai/en/wonk/productivity-and-ai. Firms need to measure cycle time, quality, error correction, and customer outcomes together. Counting generated text or model calls confuses activity with value.
A faster draft is not necessarily more output if review expands, rework increases, or downstream systems cannot accept the result. The OECD’s work on AI and productivity makes the complementary point that organizational capabilities mediate technology gains: https://oecd.ai/en/wonk/productivity-and-ai. Firms need to measure cycle time, quality, error correction, and customer outcomes together. Counting generated text or model calls confuses activity with value. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
A faster draft is not necessarily more output if review expands, rework increases, or downstream systems cannot accept the result. The OECD’s work on AI and productivity makes the complementary point that organizational capabilities mediate technology gains: https://oecd.ai/en/wonk/productivity-and-ai. Firms need to measure cycle time, quality, error correction, and customer outcomes together. Counting generated text or model calls confuses activity with value. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
The adoption curve is uneven by design
Large firms can afford data engineering, security review, and dedicated change management. Smaller firms may have stronger incentives but fewer people to integrate a tool. The World Bank’s work on AI and development provides a useful global lens: https://www.worldbank.org/en/programs/techsavvy/brief/artificial-intelligence. A global average can conceal a widening capability gap if the infrastructure and skills needed for adoption are concentrated in a few regions.
Large firms can afford data engineering, security review, and dedicated change management. Smaller firms may have stronger incentives but fewer people to integrate a tool. The World Bank’s work on AI and development provides a useful global lens: https://www.worldbank.org/en/programs/techsavvy/brief/artificial-intelligence. A global average can conceal a widening capability gap if the infrastructure and skills needed for adoption are concentrated in a few regions. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Large firms can afford data engineering, security review, and dedicated change management. Smaller firms may have stronger incentives but fewer people to integrate a tool. The World Bank’s work on AI and development provides a useful global lens: https://www.worldbank.org/en/programs/techsavvy/brief/artificial-intelligence. A global average can conceal a widening capability gap if the infrastructure and skills needed for adoption are concentrated in a few regions. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Workers experience augmentation as job redesign
Calling AI an augmentation tool does not settle what happens to a worker’s day. A support agent may handle more cases but lose discretion; an analyst may produce more scenarios but spend more time validating data. The International Labour Organization’s research on generative AI and jobs emphasizes task exposure and the importance of institutions: https://www.ilo.org/publications/generative-ai-and-jobs. The unit of change is the workflow, not the occupation label.
Calling AI an augmentation tool does not settle what happens to a worker’s day. A support agent may handle more cases but lose discretion; an analyst may produce more scenarios but spend more time validating data. The International Labour Organization’s research on generative AI and jobs emphasizes task exposure and the importance of institutions: https://www.ilo.org/publications/generative-ai-and-jobs. The unit of change is the workflow, not the occupation label. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Calling AI an augmentation tool does not settle what happens to a worker’s day. A support agent may handle more cases but lose discretion; an analyst may produce more scenarios but spend more time validating data. The International Labour Organization’s research on generative AI and jobs emphasizes task exposure and the importance of institutions: https://www.ilo.org/publications/generative-ai-and-jobs. The unit of change is the workflow, not the occupation label. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Skills are complements only when organizations invest
A model can raise the return to domain knowledge while lowering the value of routine production. That is not automatically good for workers who are not given time to acquire the new skills. The IMF’s research on skills and AI highlights the transition challenge: https://www.imf.org/en/blogs/articles/2026/01/14/new-skills-and-ai-are-reshaping-the-future-of-work. Training must include judgment, verification, data literacy, and the ability to challenge a machine output.
A model can raise the return to domain knowledge while lowering the value of routine production. That is not automatically good for workers who are not given time to acquire the new skills. The IMF’s research on skills and AI highlights the transition challenge: https://www.imf.org/en/blogs/articles/2026/01/14/new-skills-and-ai-are-reshaping-the-future-of-work. Training must include judgment, verification, data literacy, and the ability to challenge a machine output. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
A model can raise the return to domain knowledge while lowering the value of routine production. That is not automatically good for workers who are not given time to acquire the new skills. The IMF’s research on skills and AI highlights the transition challenge: https://www.imf.org/en/blogs/articles/2026/01/14/new-skills-and-ai-are-reshaping-the-future-of-work. Training must include judgment, verification, data literacy, and the ability to challenge a machine output. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Public services face a different evidence burden
A private company can experiment with a sales workflow; a public agency must justify decisions to people who may not be able to appeal an automated error. Procurement, accessibility, records retention, and due process become part of the productivity calculation. The OECD AI Principles provide a shared reference for trustworthy deployment: https://oecd.ai/en/ai-principles. A faster public process that loses legitimacy is not an efficiency gain.
A private company can experiment with a sales workflow; a public agency must justify decisions to people who may not be able to appeal an automated error. Procurement, accessibility, records retention, and due process become part of the productivity calculation. The OECD AI Principles provide a shared reference for trustworthy deployment: https://oecd.ai/en/ai-principles. A faster public process that loses legitimacy is not an efficiency gain. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
A private company can experiment with a sales workflow; a public agency must justify decisions to people who may not be able to appeal an automated error. Procurement, accessibility, records retention, and due process become part of the productivity calculation. The OECD AI Principles provide a shared reference for trustworthy deployment: https://oecd.ai/en/ai-principles. A faster public process that loses legitimacy is not an efficiency gain. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
flowchart LR
A[Claim] --> B[Deployment constraint]
B --> C[Measured outcome]
C --> D[Decision]
Capital spending can outrun organizational learning
AI investment often arrives as infrastructure, licenses, and consulting before teams have agreed on the problem. That can create a familiar pattern: high utilization in pilots, low adoption in core processes, and a growing catalogue of disconnected assistants. The IMF’s annual report places AI inside a broader macroeconomic environment where investment, energy, and financing conditions matter: https://www.imf.org/annual-report/2026/. The spending cycle needs milestones tied to realized outcomes.
AI investment often arrives as infrastructure, licenses, and consulting before teams have agreed on the problem. That can create a familiar pattern: high utilization in pilots, low adoption in core processes, and a growing catalogue of disconnected assistants. The IMF’s annual report places AI inside a broader macroeconomic environment where investment, energy, and financing conditions matter: https://www.imf.org/annual-report/2026/. The spending cycle needs milestones tied to realized outcomes. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
AI investment often arrives as infrastructure, licenses, and consulting before teams have agreed on the problem. That can create a familiar pattern: high utilization in pilots, low adoption in core processes, and a growing catalogue of disconnected assistants. The IMF’s annual report places AI inside a broader macroeconomic environment where investment, energy, and financing conditions matter: https://www.imf.org/annual-report/2026/. The spending cycle needs milestones tied to realized outcomes. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Measurement must survive the baseline problem
If a company introduces AI while demand, staffing, or product mix is changing, it becomes hard to attribute results. Randomized trials are not always possible, but staggered rollout, matched teams, and pre-registered metrics can improve the evidence. The U.S. NIST AI Risk Management Framework is useful here because it treats measurement and governance as ongoing processes: https://www.nist.gov/itl/ai-risk-management-framework. The baseline is a management asset.
If a company introduces AI while demand, staffing, or product mix is changing, it becomes hard to attribute results. Randomized trials are not always possible, but staggered rollout, matched teams, and pre-registered metrics can improve the evidence. The U.S. NIST AI Risk Management Framework is useful here because it treats measurement and governance as ongoing processes: https://www.nist.gov/itl/ai-risk-management-framework. The baseline is a management asset. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
If a company introduces AI while demand, staffing, or product mix is changing, it becomes hard to attribute results. Randomized trials are not always possible, but staggered rollout, matched teams, and pre-registered metrics can improve the evidence. The U.S. NIST AI Risk Management Framework is useful here because it treats measurement and governance as ongoing processes: https://www.nist.gov/itl/ai-risk-management-framework. The baseline is a management asset. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Disruption is also a distribution question
Even when aggregate productivity rises, the gains may accrue to firms with data, compute, and distribution. Customers may face fewer choices if AI reinforces scale advantages. The IMF’s global perspective should be paired with competition analysis and local capability building. The UN’s technology and innovation reporting offers a broader development context: https://unctad.org/topic/science-technology-and-innovation. Deployment policy has to ask who can participate, not just how fast the frontier moves.
Even when aggregate productivity rises, the gains may accrue to firms with data, compute, and distribution. Customers may face fewer choices if AI reinforces scale advantages. The IMF’s global perspective should be paired with competition analysis and local capability building. The UN’s technology and innovation reporting offers a broader development context: https://unctad.org/topic/science-technology-and-innovation. Deployment policy has to ask who can participate, not just how fast the frontier moves. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Even when aggregate productivity rises, the gains may accrue to firms with data, compute, and distribution. Customers may face fewer choices if AI reinforces scale advantages. The IMF’s global perspective should be paired with competition analysis and local capability building. The UN’s technology and innovation reporting offers a broader development context: https://unctad.org/topic/science-technology-and-innovation. Deployment policy has to ask who can participate, not just how fast the frontier moves. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Energy and infrastructure become macro variables
AI adoption has a physical footprint in data centers, networks, and electricity markets. The International Energy Agency’s electricity analysis explains why data-center demand can matter to regional grids: https://www.iea.org/reports/electricity-2026. An enterprise business case that ignores energy price, connection delays, or water and cooling constraints is incomplete. The macroeconomic effect will be shaped by where infrastructure can actually be built.
AI adoption has a physical footprint in data centers, networks, and electricity markets. The International Energy Agency’s electricity analysis explains why data-center demand can matter to regional grids: https://www.iea.org/reports/electricity-2026. An enterprise business case that ignores energy price, connection delays, or water and cooling constraints is incomplete. The macroeconomic effect will be shaped by where infrastructure can actually be built. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
AI adoption has a physical footprint in data centers, networks, and electricity markets. The International Energy Agency’s electricity analysis explains why data-center demand can matter to regional grids: https://www.iea.org/reports/electricity-2026. An enterprise business case that ignores energy price, connection delays, or water and cooling constraints is incomplete. The macroeconomic effect will be shaped by where infrastructure can actually be built. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
The best early metric is verified task completion
Organizations need a small set of measures that connect model use to a human or customer outcome: accepted cases, safe resolutions, time to decision, repeat work, and escalation quality. This is less glamorous than tracking tokens, but it answers the economic question. A model can be impressive and still fail the process if humans must inspect every output with no net time saved.
Organizations need a small set of measures that connect model use to a human or customer outcome: accepted cases, safe resolutions, time to decision, repeat work, and escalation quality. This is less glamorous than tracking tokens, but it answers the economic question. A model can be impressive and still fail the process if humans must inspect every output with no net time saved. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
Organizations need a small set of measures that connect model use to a human or customer outcome: accepted cases, safe resolutions, time to decision, repeat work, and escalation quality. This is less glamorous than tracking tokens, but it answers the economic question. A model can be impressive and still fail the process if humans must inspect every output with no net time saved. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
The evidence clock is now running
The IMF report should be treated as a measurement agenda. Over the next year, firms should publish clearer baselines, workers should be included in redesign decisions, and governments should track adoption by sector rather than relying on vendor counts. AI will change work, but the size and distribution of that change are not pre-written. They will be determined by the quality of the institutions that deploy it.
The IMF report should be treated as a measurement agenda. Over the next year, firms should publish clearer baselines, workers should be included in redesign decisions, and governments should track adoption by sector rather than relying on vendor counts. AI will change work, but the size and distribution of that change are not pre-written. They will be determined by the quality of the institutions that deploy it. The first reporting question is not whether the idea sounds ambitious, but which dependency could make it fail in an ordinary production environment. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
The IMF report should be treated as a measurement agenda. Over the next year, firms should publish clearer baselines, workers should be included in redesign decisions, and governments should track adoption by sector rather than relying on vendor counts. AI will change work, but the size and distribution of that change are not pre-written. They will be determined by the quality of the institutions that deploy it. A second question concerns incentives. The organization making the announcement benefits from a favorable interpretation, while customers, workers, or operators carry different risks. Good analysis keeps those positions visible and asks what independent measurement would change the conclusion. The practical implication is that readers should separate what the source directly establishes from what follows as analysis. That discipline matters for the imf’s ai deployment report puts productivity claims on an evidence clock: a plausible mechanism is not the same thing as observed performance, and a vendor, research group, or forecaster should be credited for its claim rather than treated as a neutral narrator.
The reporting standard that matters
The strongest way to follow this story is to watch the evidence trail rather than the announcement cycle. Primary documents establish what was released, promised, or measured. Independent evaluations establish whether the result generalizes. Operational data establishes whether people can use it repeatedly. Those are different forms of evidence, and collapsing them into one headline is how AI coverage becomes less useful than the systems it describes.
Sources consulted
- https://www.imf.org/annual-report/2026/in-focus/ai-deployment-and-disruption/
- https://www.imf.org/annual-report/2026/
- https://www.oecd.ai/en/wonk/productivity-and-ai
- https://www.worldbank.org/en/programs/techsavvy/brief/artificial-intelligence
- https://www.ilo.org/publications/generative-ai-and-jobs
- https://www.imf.org/en/blogs/articles/2026/01/14/new-skills-and-ai-are-reshaping-the-future-of-work
- https://oecd.ai/en/ai-principles
- https://www.nist.gov/itl/ai-risk-management-framework
- https://unctad.org/topic/science-technology-and-innovation
- https://www.iea.org/reports/electricity-2026