California’s SB 253 sets a phased schedule: Scope 1 and 2 emissions in the first reporting year, followed by Scope 3 in 2027.
Confirm current filing dates and submission requirements with the California Air Resources Board, because implementation details keep moving. AI can reduce repetitive data handling. Reliable reporting still depends on clear methods, source records, and human review.
Key Takeaways
- Scope of the law. SB 253 covers US-based business entities doing business in California with more than $1 billion in annual revenue.
- The first phase covers Scope 1 and 2. Check current CARB instructions for deadlines, formats, and assurance requirements.
- The reporting year is usually FY2025. Deadline extensions have not changed which year’s data is due.
- Audit readiness requires traceability. Every figure should connect to a source document, an emission factor, a calculation method, and a change history.
- AI has a defined role. It helps with intake, factor matching, quality checks, and evidence organisation. People set methodology and approve results.
What SB 253 Requires
Under California Health and Safety Code Section 38532, as amended by SB 219, covered entities must disclose greenhouse gas emissions annually using the GHG Protocol.
Scope 1 covers direct emissions from owned or controlled sources. Scope 2 covers emissions from purchased energy. And, Scope 3 covers other indirect emissions across the value chain.
The statute calls for limited assurance of Scope 1 and 2 beginning in the first reporting year, moving to reasonable assurance in 2030. Limited assurance involves less extensive procedures than reasonable assurance.
For the first reports, CARB has indicated that limited assurance will not be required, exercising enforcement discretion and accepting Scope 1 and 2 data either way. For Scope 3, the statute allows CARB to set an assurance requirement by January 1, 2027, with limited assurance beginning in 2030.
Distinguish statutory requirements from current implementation guidance. Confirm applicable obligations, transitional provisions, and interpretation questions with CARB and counsel.
Which Year’s Data Are You Reporting?
This trips up more teams than the deadline itself.
The reporting year follows your fiscal year-end rather than the submission date. Entities whose fiscal year ends between 2 February and 31 December 2026 report data from the fiscal year ending in 2025. Entities whose fiscal year ends between 1 January and 1 February 2026 report data from the fiscal year ending in 2026.
Deadline extensions have not changed this. Additional time validates the same data rather than moving the period forward.
Audit Readiness, Translated
An assurance provider needs to trace reported figures back through the records behind them. Each number should connect to its source document, emission factor, calculation method, organisational boundary and a record of who changed what.
An emission factor converts an activity, such as electricity use, into an emissions estimate.
The minimum evidence set includes:
- Source documents such as utility bills, fuel invoices and meter exports
- Factor details, including dataset, version, geography and year
- Calculation records that allow results to be reproduced
- Approvals with names and timestamps
- Change history for data, factors and methods
- Roles and permissions records
Build these records while collecting data. Reconstructing an audit trail afterward delays reporting and leaves gaps that are difficult to resolve.
There is a second reason to document as you go. CARB has described a first-year enforcement posture that rewards demonstrable good-faith effort rather than perfect numbers, and demonstrable effort means a written record of data sources, assumptions, known gaps, and a plan to close them.
Where AI Helps and Where It Doesn’t
AI speeds up repetitive handling and review. People remain responsible for methodology, reporting boundaries, and sign-off.
1. Intake and Normalisation
Document tools can extract quantities and units from bills and invoices, then suggest activity categories. Duplicate checks and unit conversions prepare records for calculation.
Compare a sample of extracted data against the originals, and review ambiguous fields before accepting them.
2. Activity-to-Factor Matching
AI can propose emission factors for activity records and flag uncertain matches. Treat a confidence label as a review aid rather than proof of accuracy.
Require a documented reason for each match and a way for reviewers to reject or correct it.
3. Calculations
Keep calculations rule-based, using documented factor versions. AI can flag when a newer factor exists. It should never replace an approved factor silently.
Record and approve updates so previous results stay reproducible.
4. Supplier Engagement for Scope 3
Automated requests, submitted-data checks and status tracking help teams manage outreach at scale.
Keep supplier-reported measurements separate from estimates, and record the method behind each figure.
5. Quality Checks
AI can flag unusual changes for investigation. Reconcile utility totals against meter or finance records, and document how each flag was resolved.
An unusual result may reflect a real operational change rather than an error.
Not every step needs AI. Standard rules handle calculation and validation well. AI earns its place where documents vary in format or records need suggested classification.
Data Governance for SB 253
Document these controls in an internal procedure and assign an owner to each:
- One emissions ledger with a permanent, unique ID for each line item
- Factor details stored with every calculation
- Access permissions based on job responsibilities, with defined approval steps
- Saved versions of data and calculations so reported figures can be reproduced
- A consistent organisational boundary, based on ownership share or control
- A record-retention plan supporting reporting and assurance needs
Sweep, the sustainability intelligence platform, addresses several of these controls in one system. It centralises emissions data collection, calculation and reporting, keeps each calculation and methodology traceable for CARB and assurance providers, and supports Scope 3 supplier outreach through surveys with automated reminders.
Whichever system holds your ledger, confirm it supports every control listed above.
Preparing for the 2026 Filing
Build your plan around the Scope 1 and 2 start date, then verify CARB’s current instructions.
As of September 2026, CARB has proposed moving the first-year deadline from August 10 to November 10, 2026. That change came on June 24, 2026, after CARB withdrew its initial regulation from the Office of Administrative Law for revision. The revised regulation still requires OAL approval following a 15-day public comment period.
Plan against November 10 while monitoring CARB’s rulemaking page. CARB has already moved this date more than once.
Before submission, confirm:
- The applicable deadline and reporting period
- Required forms, file formats, and submission channel
- Assurance requirements and any first-year transitional guidance
Keep methodology notes and a boundary description ready alongside emissions totals. Preserve supporting evidence even where the submission process does not require every underlying document.
Track SB 253 and SB 261 Separately
These laws are moving on different tracks, and conflating them causes real planning errors.
On November 18, 2025, the Ninth Circuit granted an injunction pausing enforcement of SB 261, while expressly declining to enjoin SB 253. Oral argument was heard on January 9, 2026, and a merits decision is awaited.
SB 253 remains in effect and implementation continues. SB 261 reporting is voluntary until the Ninth Circuit rules, and CARB will set an alternate date once the appeal resolves.
The practical consequence: SB 261’s legal uncertainty should not create a sense that SB 253 is similarly unsettled. It is not. Meanwhile, companies in scope for SB 261 should keep their climate risk work moving, because a dissolved injunction could leave little preparation time.
A Sample 45-Day Evidence Plan
This sequence suits a team with source records already available. Allow more time where records are incomplete, several entities are involved, or external review requires a longer schedule.
Days 1–7. Confirm reporting entities, boundary method, and data sources, including utilities, fuel, refrigerants, and fleet records. Assign owners.
Days 8–14. Import documents and use AI extraction where helpful. Remove duplicates, standardise units, record gaps.
Days 15–24. Match activities to factors. Review uncertain matches and document decisions.
Days 25–33. Reconcile utility totals against meter and finance data. Resolve unusual results with written notes.
Days 34–40. Assemble source documents, factor details, calculation records, approvals and change history.
Days 41–45. Complete internal review, repeat a sample of calculations, obtain sign-off. Coordinate any external assurance separately.
The resulting evidence pack should contain emissions totals by scope, a methodology memo, a boundary statement, a factor register, an evidence index and an approval record.
What to Ask AI-Enabled Vendors
Request a demonstration or supporting document for each capability:
- Factor transparency. Show one factor’s source and version history, including how an approved version is retained.
- Review controls. Demonstrate an uncertain factor match and how a reviewer accepts, rejects or changes it.
- Traceable calculations. Export one line item from the source document through to the final result.
- Security. Provide relevant SOC 2 reports or ISO 27001 certification, data-location details, and terms governing AI use of uploaded data.
- Approval steps. Show where a missing approval prevents a record from being finalised.
- Supplier data checks. Submit an invalid value and demonstrate how the system handles it.
- Evidence export. Produce a sample pack an independent reviewer can navigate.
Test the workflow with representative records from your own business, and confirm evidence remains accessible outside the software.
AI Pitfalls to Avoid
Watch for silent methodology changes, invented emission factors, misclassified activities, and sensitive data sent to unapproved models.
That last point deserves attention beyond emissions work. Whether a given document is safe to paste into a chatbot needs a written answer before anyone faces a deadline, since operational and financial data flows through this process.
Restrict factor selection to approved libraries, document matching decisions, enforce approval steps, and repeat a sample of calculations before filing. The discipline resembles what AI-assisted bookkeeping demands: automation accelerates the work while verification determines whether the output holds.
Scope 3 Planning for 2027
Start by screening all 15 GHG Protocol Scope 3 categories for relevance.
Prioritise better data where expected emissions and business relevance are greatest, rather than relying on supplier spend alone. Seek supplier-specific data where useful, and document estimates where direct data is unavailable.
Build emissions-data requirements into new supplier contracts, and pilot collection in relevant categories such as purchased goods and services, fuel- and energy-related activities, waste, business travel and employee commuting. The most important categories depend on the business.
Scope 3 typically represents the largest share of a company’s footprint and the hardest data to defend under scrutiny. Starting now is the difference between an inventory and a scramble.
The Bottom Line
Prepare Scope 1 and 2 totals with traceable evidence, and confirm the applicable CARB deadline and filing requirements. In parallel, build the Scope 3 inventory and supplier process.
Let AI assist with extraction, suggested factor matches, unusual-result checks, and evidence organisation, while people set methods and approve results.
Whether the ledger lives in a platform or an internal system, the test stays the same: can an independent reviewer trace and reproduce each reported number?
Related: How Swiss Data Protection Affects AI Companies vs EU GDPR
