AI Risk Movers & Shakers - September 2026

24 September 2026 — Baseline Updates, Site Changes

By Craig Atkins, Founder, AI Risks

Published 24 September 2026; updated 24 September 2026; latest published AI Risks data release 2026-09-23.

Our September 2026 baseline update changes scores for 83 existing entries and adds seven emerging sectors. The sharpest movements show growing pressure on routine digital work, alongside new opportunities in agent security, evaluation, robotics, infrastructure and commerce.

The question running through our September update is no longer simply whether AI can help someone do a task. It is what happens when AI can carry out more of the workflow around that task as well. That shift affects a software business, an accountancy practice and a marketing agency rather differently, but it gives all three a reason to revisit their assumptions.

We have rebased the AI Risks sector library to reflect faster adoption of agentic AI, coding automation, multimodal generation and increasingly capable physical automation. Of the 145 existing entries reviewed, 83 have a changed opportunity or resilience score. We have also added seven emerging industry sectors, taking the library to 138 industry sectors and 14 broader macro sectors. These are changes to our assessment of the outlook, not a claim that 83 industries transformed in a month.

The biggest move: routine digital work is under more pressure

Professional Services moves from 5/4/3/2 to 4/3/2/1 for resilience across the 3, 5, 7 and 10-year horizons, while its AI opportunity score rises from 8 to 9. Lower resilience means greater estimated exposure to disruption. The opportunity and the risk are moving together because research, drafting, analysis and administrative execution can increasingly be joined into a single workflow.

Within that group, Management Consulting moves from 5/4/3/2 to 4/3/2/1, with opportunity rising from 9 to 10. Translation and language services move from 4/3/2/1 to 3/2/1/1, also reaching 10 for opportunity. Accounting, bookkeeping and tax services fall from 4/3/2/1 to 3/2/1/1 on resilience, although their opportunity score remains at 9.

This does not mean clients will stop paying for judgement, accountability or a person who understands the awkward detail of their business. It does mean that charging for routine production alone looks less comfortable. The question for a professional firm is which part of its service the client will still value when a competent first draft becomes cheap and quick.

Software faces a similar test. Technology & Software moves from 6/5/4/3 to 5/4/3/2 for resilience, with opportunity holding at 10. Software companies selling products make the same resilience move. Coding agents can shorten development and testing cycles for vendors, but they can also make it easier for customers and competitors to build features that once justified a separate product. Deep integration, proprietary data, reliability and distribution matter more when producing code is less scarce.

Two sectors highlighted in August's Movers & Shakers move again. Online Tutoring / EdTech goes from 3/2/1/1 to 2/1/1/1 for resilience, and Video Production / Creative Studios from 4/3/2/1 to 3/2/1/1. Both retain an opportunity score of 10. The scope to deliver more at lower cost is considerable; so is the pressure on businesses selling a standardised version of what AI can now produce.

One of the most striking industry moves is Data Annotation / AI Evaluation Services. Its resilience falls from 5/4/3/2 to 3/2/1/1, while opportunity rises from 8 to 10. That mixed category contains two very different futures: commodity labelling is increasingly exposed, while expert evaluation and difficult domain-specific testing may become more valuable. It is a useful reminder that a sector label is a starting point, not a verdict on every business inside it.

Seven new sectors show where the next layer of work is forming

This month we added AI Security / Agent Identity & Access Management, AI Agent Orchestration / Enterprise Automation Platforms, and AI Evaluation / Red Teaming Laboratories. Together they reflect a practical problem: once agents can act across systems, organisations need to decide what those agents may access, how their work is coordinated, and how to test whether they behave as intended. Each starts with an opportunity score of 10 in the new baseline.

We also added Synthetic Media Provenance / Content Authenticity, Humanoid / General-Purpose Robot Fleet Operators, AI Data Centre Cooling / Power Optimisation, and Agentic Commerce / AI Shopping Optimisation. These are quite different businesses, but each supplies something that more capable AI will need: trust in digital content, deployment of physical systems, power and cooling for compute, or a way for merchants to be found when shopping agents mediate the purchase.

A new category is not automatically a safe investment. Agent orchestration, for example, has an opportunity score of 10 but resilience declines from 8 at three years to 4 at ten years. Large platforms may absorb some of the functions that specialist providers currently sell. The same question applies to any emerging market: what will remain distinctive once the obvious features become standard?

The physical world still complicates the story

Not every September adjustment points towards lower resilience. Taxi / Private Hire rises from 5/4/3/2 to 7/6/5/4. AI opportunity remains at 9, but our rebase gives more weight to the practical and regulatory barriers to replacing real-world service at scale. AI Governance / Assurance / Model Risk and Robotics Integrator / Automation Engineering also gain a point of resilience across all four horizons, with opportunity holding at 10.

For IT Managed Service Providers, the resilience profile stays at 6/5/4/3 while opportunity rises from 9 to 10. That is a useful distinction for MSPs and their clients. There may be more scope to help a business apply AI, secure it and change workflows, even where the near-term exposure of the MSP model itself has not materially changed in this update.

What should you do with a changed score?

Start with a client or prospect whose revenue depends on repeatable digital work. Ask which part of the service could now be produced, delivered or bought differently, and which part still depends on trust, specialist knowledge or real-world execution. Then look at the opportunity side: could the business use the same capabilities to improve its own delivery before somebody else does?

Our resilience scores are sector benchmarks, not forecasts for individual companies. A lower score should prompt a better question, not a dramatic board paper on its own. The AI Risks methodology explains how we use public information, sector context and time horizons. You can explore the AI Disruption Index to compare September with previous releases.

The September lesson is that opportunity is not reserved for the sectors with the strongest defences. In several of the most exposed industries, the opportunity score is also at or near the top of the scale. The interesting question is whether a business uses that opportunity to change its proposition, or waits for a competitor to make the introduction.