UK climate reporting in 2026 is not one universal checklist. The duties that apply depend on whether an organisation is listed, falls within the Companies Act climate-disclosure scope, reports under Streamlined Energy and Carbon Reporting (SECR), or chooses a voluntary framework. AI can help reconcile evidence and expose inconsistencies, but it cannot determine legal scope, materiality or what directors should say.
This guide reflects the position in the United Kingdom on 31 July 2026. It is operational guidance, not legal, accounting or assurance advice. Teams should confirm the rules for their entity, group structure, financial year and sector with qualified advisers.
What changed, and what did not
The government issued UK Sustainability Reporting Standards S1 and S2 on 25 February 2026. S1 covers general sustainability-related financial information; S2 covers climate-related risks and opportunities. Both are based on the ISSB baseline with UK amendments.
Their arrival did not make them mandatory for every UK company. The FRC’s February 2026 sustainability FAQs state that UK SRS are available for voluntary use immediately and are not currently compulsory. The FCA’s consultation proposed UK-SRS-aligned rules from 1 January 2027, but at this article’s cutoff the consultation had closed and no final policy statement had been published. The FCA reporting-requirements page/reporting-requirements) remains the current source for listed issuers and regulated firms.
Existing obligations continue. In-scope companies may need a Non-Financial and Sustainability Information Statement with climate-related financial disclosures; quoted and qualifying large entities may also face SECR energy and emissions requirements. The government’s 2026 SECR post-implementation review material confirms that the framework remains mandatory for quoted companies, large unquoted companies and qualifying LLPs.
| Reporting route | Position at 31 July 2026 | AI’s legitimate role |
|---|---|---|
| UK SRS S1/S2 | Voluntary unless another requirement or commitment makes it relevant | Map evidence, test consistency and draft traceable working text |
| Companies Act climate disclosures | Mandatory for entities within the statutory scope | Assemble source records; never decide scope or materiality |
| FCA TCFD-aligned rules | Current rules continue for relevant listed issuers and firms | Compare claims, metrics and risk-register entries |
| SECR | Continues separately for qualifying entities | Reconcile energy data, factors, boundaries and prior-year movements |
| Customer or lender requests | Contract-specific | Produce controlled responses from approved facts |
Start with an applicability register
A finance team should not ask a model, “What must we disclose?” and accept its answer. Create a signed applicability register maintained by the company secretary, finance lead and legal adviser. Record the legal entity, listing status, turnover, employee and balance-sheet tests, group exemptions, financial year, current reporting route, voluntary commitments and accountable owner.
Attach the authoritative source and the date it was checked to every conclusion. Separate enacted requirements from consultations, market expectations and internal ambitions. Reconfirm the register when an acquisition, refinancing, listing change or year-end alters scope.
The FRC explains that voluntary use of UK SRS S2 can avoid duplicating Companies Act climate disclosures only when its use is clearly referenced in the NFSIS and the remaining statutory conditions are met. It also says SECR data currently remains separate. That nuance is exactly where a generic AI summary can create a false compliance claim.
Build a controlled evidence spine
Reliable reporting begins below the narrative. Create one evidence register that links each reported statement to a source, method, owner and approval. The register should cover facilities, energy invoices, fleet records, purchased goods, investment data where relevant, climate scenarios, insurance information, operational dependencies and financial-planning assumptions.
For every metric, capture:
- reporting entity and operational or financial boundary;
- period, units, currency and conversion rules;
- original system, file and source owner;
- emission factor, version and factor year;
- estimation method and reason for its use;
- recalculation and prior-year-restatement policy;
- review status, approver and immutable timestamp;
- disclosure paragraph, table or target that consumes the value.
Use deterministic software for arithmetic and transformations. An AI assistant can classify invoices, suggest mappings and explain anomalies, but the approved calculation should run in version-controlled code or a governed reporting platform. Retain source snapshots and hashes so reviewers can reproduce the number later.
Teams beginning this foundation may also use our guide to sustainable AI and data infrastructure and the UK smart-grid and energy AI operating guide to connect corporate reporting with operational data.
Put AI around the ledger, not in place of it
Useful climate-reporting applications are bounded and reviewable. Retrieval can find the passages in board papers that support a risk statement. Classification can route invoices to an energy category. Anomaly detection can highlight an improbable intensity change. A language model can compare the annual report, website, lender questionnaire and procurement response for inconsistent dates or targets.
The workflow should be evidence first:
- Ingest only approved, access-controlled source material.
- Extract candidate facts with page, cell or record references.
- Validate quantities, dates and entities using deterministic rules.
- Route exceptions to the named data owner.
- Generate draft language only from accepted facts.
- Require finance, sustainability, legal and disclosure-committee review.
- Freeze the evidence pack, prompt version, model version and approvals.
- Re-run consistency checks against every public channel before release.
Do not let the model invent missing values, select a favourable scenario or convert an aspirational plan into a target. If evidence is incomplete, the output should display an explicit gap and confidence state, not fluent filler.
Treat narrative claims as controlled data
Climate reporting fails when numbers and prose travel through different approval routes. Store each material claim as a structured record: text, claim type, applicable entity, period, source, uncertainty, owner, status and public destinations. That makes it possible to identify a website promise that no longer matches the annual report.
The FRC Guidance on the Strategic Report should inform how the business connects principal risks, strategy, business model, performance and prospects. AI may reveal missing connections, but directors retain responsibility for a fair, balanced and understandable report.
Consumer-facing statements require a separate test. Under the CMA’s unfair commercial practices guidance, environmental claims must not mislead through false wording or omitted context. A reporting-compliant metric is not automatically an acceptable advert. Record the population, baseline, exclusions and evidence behind terms such as “net zero”, “renewable” or “low carbon”.
Privacy, security and supplier controls
Climate evidence can reveal site output, supplier pricing, property details, travel patterns and strategic vulnerabilities. Personal data may appear in expenses, vehicle logs or supplier contacts. Apply purpose limitation, minimisation, retention and access controls described in our UK AI and data-protection guide.
Before connecting a model, complete a data-flow map and vendor assessment. Establish where prompts, files and outputs are stored; whether the provider trains on them; which subprocessors and transfer mechanisms apply; how deletion works; and whether the business can export a complete audit trail.
Follow the NCSC’s secure AI system development guidelines, including asset inventories, supply-chain controls, logging and restoration to a known-good state. The government’s AI Cyber Security Code of Practice supplies a further baseline for AI providers and deployers.
Access should be role-based. A drafting assistant does not need write access to the emissions ledger, and an external assurance team should receive a controlled evidence room rather than unrestricted model access. Test prompt injection in supplier documents, hidden spreadsheet content, malicious links and unauthorised cross-entity retrieval.
Governance and assurance that can survive challenge
Assign one accountable executive, but distribute control ownership. Finance owns reported numbers and reconciliation; sustainability owns methodology and domain interpretation; legal and company secretariat own applicability and statements; security and privacy own system controls; internal audit tests design and operation. The board or relevant committee approves material judgements.
Maintain an AI use-case register with purpose, owner, model, data classes, permitted actions, evaluation results and withdrawal procedure. Label model-produced text in the working papers. A reviewer should see the supporting evidence beside every material sentence, not search through a chat history.
Evaluate against a fixed, representative set that includes awkward cases: acquired sites, partial invoices, renewable certificates, estimation gaps, factor changes, overseas operations and contradictory source documents. Measure factual precision and missed exceptions separately. A polished paragraph with one unsupported figure is a failure.
A measurable 90-day implementation
Days 1–30 — scope and baseline. Sign the applicability register; inventory disclosures, systems and public claims; choose one reporting entity and ten high-risk metrics; map evidence lineage; complete privacy and threat assessments. Baseline reconciliation time, unresolved exceptions, manual hand-offs and unsupported claims.
Days 31–60 — controlled pilot. Run extraction and anomaly detection on a copy of approved data. Build deterministic calculation checks and a claim register. Test model and prompt changes against the fixed evaluation set. Give reviewers side-by-side evidence and capture corrections by failure type.
Days 61–90 — parallel close. Operate the new process beside the existing close without replacing it. Reconcile results, test access and deletion, conduct an incident exercise, and ask internal audit or an independent reviewer to sample lineage. The disclosure committee then decides whether to expand.
Proceed only if the pilot achieves all of these gates:
- 100% of material metrics resolve to approved evidence and methodology;
- zero unreviewed model text enters a board or public document;
- at least 95% precision on extracted fields, with every miss classified;
- all boundary, factor and restatement changes produce visible exceptions;
- no critical access, retention or cross-entity leakage finding remains open;
- reviewer time falls without increasing post-review corrections.
Pause if the model silently fills missing data, cannot reproduce an output, changes a target’s meaning, exposes restricted records, or performs worse for a material business unit. Also pause when a vendor update changes behaviour before regression testing, or when scope advice is unresolved. The valuable outcome is not more climate prose; it is a shorter, traceable path from operational evidence to an accountable decision.
Primary sources checked
- UK SRS S1 and S2, Department for Business and Trade, 25 February 2026
- Sustainability Reporting Developments FAQs, FRC, updated 26 February 2026
- Sustainability reporting requirements, FCA
- Guidance on the Strategic Report, FRC, revised February 2026
- SECR post-implementation review opinion, Regulatory Policy Committee, June 2026
- Unfair commercial practices, CMA
- Guidelines for secure AI system development, NCSC
- AI Cyber Security Code of Practice, DSIT, 31 January 2025
