An AI-generated cosmetic formula can pass a Cosmetic Product Safety Report (CPSR) in both the EU and UK but only after it has been reviewed, tested, and approved by a qualified cosmetic safety assessor. AI can speed up formulation, but it cannot replace toxicological assessment, laboratory testing, or the legal responsibilities required under cosmetic law.
AI formulation tools are becoming common in cosmetic development, but regulators don’t approve formulas based on how they were created.
They approve them based on safety data, documentation, and laboratory testing. An AI cosmetic formula CPSR still has to clear the same bar as one built by a human chemist.
Cosmetic AI compliance isn’t about banning AI from formulation. It’s about knowing where AI’s job ends, and the legal safety assessment begins. Under EU Regulation 1223/2009, every product placed on the UK or EU market needs a CPSR before sale, and that doesn’t bend for AI-assisted development.
This guide covers:
- What happens during an AI CPSR assessment
- Why AI formula safety EU standards require human toxicological review
- A real-world example of an AI formula moving through CPSR review
- How AI beauty formulation UK tools work, and their limits
- The documentation and Margin of Safety (MoS) calculations assessors check
- A step-by-step answer to how to get a CPSR for an AI-generated cosmetic formula in the UK
If you’re a brand owner, formulator, or seller using AI to speed up development, here’s what you still need to do before that formula can legally go to market.
Can AI-generated cosmetic formulas pass a CPSR in the EU and UK?
Yes, but only after a qualified safety assessment. An AI-generated formula can pass a CPSR if it meets the same ingredient, concentration, and documentation standards as any other formula. The AI origin is irrelevant to the assessor.
What matters is whether the ingredients are legal at the percentages used and whether the product is safe under realistic use.
What Safety Assessors Actually Evaluate
A qualified safety assessor looks at the finished formula, not the process that created it. They:
- Check ingredient legality against Annexes II to VI
- Calculate exposure and Margin of Safety
- Confirm the product won’t cause harm under normal use
None of this depends on who, or what, wrote the formula.
Why AI Origin Doesn’t Affect Approval Outcome
Regulators have no separate category or pathway for “AI-generated” formulas under Regulation 1223/2009. A formula either meets the legal and toxicological requirements, or it doesn’t. That said, AI tools tend to introduce specific, recurring problems that human formulators are less likely to make.
When AI Cosmetic Formulas Fail CPSR Review
Most failures trace back to restricted ingredients, concentrations exceeding SCCS limits, and formulas built from training data that misses recent regulatory changes. AI models often have a training cutoff, so a tool might suggest a concentration that was fine two years ago but is now restricted.
What AI Cosmetic Formulation Software Can Do
AI can make the formulation process faster by using data from thousands of cosmetic ingredients. It can help you:
- Recommend ingredients for your product
- Suggest ingredient percentages
- Compare ingredient databases
- Estimate formulation costs
- Predict how stable a formula may be
- Improve texture and product feel
- Suggest replacement ingredients if one is unavailable
AI is a helpful tool for planning and improving formulas, but its suggestions are based on data, not real testing.
What AI Cannot Do
There are some things AI cannot replace. It cannot:
- Carry out laboratory testing
- Perform toxicology assessments
- Create a Cosmetic Product Safety Report (CPSR)
- Approve your product for sale
- Give the final safety sign-off
AI can help you build a formula, but every cosmetic product still needs proper testing, safety checks, and regulatory approval before it goes on the market.
What Happens During an AI CPSR Assessment?
Step-by-Step AI CPSR Assessment Process
The process is the same whether the formula comes from AI or a human chemist, but AI formulas often need more corrections at the start.
- Review the ingredient list against current Annex restrictions
- Identify and correct any non-compliant or restricted ingredients
- Compile all supporting documentation (SDS, TDS, CoA)
- Run exposure modelling for each ingredient
- Calculate the Margin of Safety (MoS)
- Write the safety conclusion for CPSR Part B
What Toxicologists Check in Practice
Toxicologists look for specific red flags during review, including:
- Ingredients used above their allowed concentration
- Vague INCI naming that doesn’t match an approved substance
- Active combinations that have never been tested together
- Whether the AI tool accounted for cumulative exposure, since most tools generate single-ingredient suggestions without modelling the full formula as a system
Why AI Output Alone Cannot Be Approved
An AI tool can suggest a formula, but it can’t sign a legal document. Under Article 10 of Regulation 1223/2009, the CPSR must be prepared and signed off by a suitably qualified Safety Assessor with a recognised qualification in pharmacy, toxicology, or medicine. AI output has no legal standing on its own.
| Task | AI Formulation Tool | Qualified Safety Assessor |
|---|---|---|
| Suggesting ingredient combinations | Yes | Yes |
| Checking Annex II–VI compliance | Partial, often outdated | Yes |
| Calculating Margin of Safety (MoS) | No | Yes |
| Reviewing supplier documentation | No | Yes |
| Signing CPSR Part B | No (not legally permitted) | Yes |
| Submitting to CPNP/SCPN | No | Yes (via Responsible Person) |
AI Cosmetic Formula Review: From AI Output to CPSR Approval
AI Skincare Formula Generated
Imagine a brand using an AI tool to build a brightening serum. The AI suggests niacinamide at 10%, a vitamin C derivative, hyaluronic acid, and a fragrance blend for a “spa-like scent,” with a preservative pulled from a generic database.
AI CPSR Compliance Issues
Compliance Issues and Required Changes
On review, the assessor found three problems:
- Undeclared fragrance allergens exceeding the Annex III labelling threshold
- A preservative is no longer compliant at the suggested concentration, since its limit was revised
- No stability data showing the vitamin C and niacinamide combination won’t destabilise over the shelf life
Fixing this means reformulating the preservative system, adding allergen labelling, and commissioning stability testing. None of it means starting over. It means correcting what the AI didn’t catch.
Final CPSR Approval Outcome
Once corrections are made, the assessor recalculates the Margin of Safety and confirms compliance. The formula passes, but only after human review catches what the AI missed. This is the pattern most AI-assisted formulas follow: promising start, a few fixable gaps, then approval.
How AI Beauty Formulation UK Tools Work in 2026
Modern AI formulation platforms match desired outcomes, like “hydrating” or “anti-ageing,” against ingredient databases and predict compatible combinations. Some models texture or estimate the per-unit cost. AI speeds up early development, generating dozens of variations in the time it takes a human formulator to draft one. But the moment compliance enters the picture, the gap shows up. Most AI tools aren’t connected to live regulatory databases, can’t calculate Margin of Safety, and have no mechanism for legal sign-off.
| Stage | What AI Handles Well | Where It Falls Short |
|---|---|---|
| Ideation | Generates variations fast | N/A |
| Cost estimation | Rough per-unit pricing | Doesn’t factor in compliance costs |
| Regulatory compliance | Limited, often outdated | Can’t flag recent Annex/SCCS updates |
| Safety calculation | None | Cannot calculate MoS |
| Legal sign-off | None | No mechanism at all |
Cosmetic AI Compliance Under EU Regulation 1223/2009
Understanding the EU and UK Frameworks
EU Regulation 1223/2009 is the core legislation governing cosmetics sold across the EU, setting requirements for safety assessment, labelling, and market responsibility.
Following Brexit, the UK adopted its own version through the UK Cosmetic Regulation 2020 (SI 2020/1466), mirroring the EU framework but running as a separate legal system with its own notification portal.
A product compliant in the EU isn’t automatically compliant in the UK, even if it has an identical formula.
Responsible Person Duties and Why AI Can’t Replace Them
Every cosmetic product needs a named Responsible Person who is legally accountable for compliance. They keep the CPSR alive, keep the PIF available, and respond to enforcement bodies like the OPSS. No AI tool can hold this responsibility. If a product causes harm and its formula was generated by an unassessed AI tool, the Responsible Person answers to regulators, not the software.
| Compliance Task | AI Tool Capability | Human Responsibility (Legal) |
|---|---|---|
| Generating formula concepts | Yes | N/A |
| Holding legal liability | No | Yes (Responsible Person) |
| Signing safety assessments | No | Yes (qualified assessor) |
| Maintaining the PIF | No | Yes |
| Responding to OPSS/enforcement | No | Yes |
Why AI Formula Safety EU Requires Human Validation
A regulatory safety assessment draws on toxicological data and judgment built from years of training. AI tools can surface relevant data points, but can’t weigh conflicting evidence or take legal accountability for the conclusion.
Most train on datasets with a fixed cutoff, so they may miss SCCS opinions or restrictions published just before launch. AI tools have suggested ingredients at concentrations that were once standard but are now restricted, or fragrance combinations that trigger allergen labelling without flagging it.
The recurring failures are rarely exotic:
- Preservatives used above current limits
- Missing allergen declarations
- Active combinations with no stability data behind them
CPSR Explained: CPSR Part A & Part B
Part A is the safety information section, covering formula composition, physical and chemical characteristics, packaging material, and the toxicological profile of each substance.
Part B is the safety assessment conclusion, in which the qualified Safety Assessor reviews Part A and provides professional judgment on whether the product is safe, including any required label warnings.
Only someone with a recognised qualification in pharmacy, toxicology, or medicine, as defined under Article 10, can sign Part B.
| Element | CPSR Part A | CPSR Part B |
|---|---|---|
| Content type | Raw data and documentation | Professional safety conclusion |
| Who compiles it | Formulator or supporting team | Qualified Safety Assessor |
| Includes toxicological conclusion | No | Yes |
| Legal sign-off required | No | Yes |
| Can AI contribute | Partially (data gathering) | No |
Product Information File (PIF) and Documentation Requirements
The PIF holds the full compliance record: the CPSR, product description, manufacturing method, and labelling details. Without a signed CPSR, the PIF is incomplete, and the product isn’t legally ready for sale. Beyond the CPSR, assessors need:
- Safety Data Sheets for every raw material
- Technical Data Sheets for every raw material
- Certificates of Analysis confirming batch purity
- IFRA or allergen declarations for fragrance components
AI tools don’t generate or source any of this. It comes directly from suppliers, and gathering it is one of the most common gaps in AI-assisted formulas.
Margin of Safety (MoS) in Cosmetic Safety Assessment
Margin of Safety (MoS) is a key part of cosmetic safety assessment used in CPSR under EU and UK regulations. It compares the safe dose of an ingredient from toxicology studies with the amount a consumer is actually exposed to during normal product use. A higher MoS means the ingredient has a larger safety buffer and is considered lower risk.
MoS = NOAEL ÷ Systemic Exposure Dose (SED)
Where:
- NOAEL (No Observed Adverse Effect Level) is the highest dose that shows no harmful effects in studies
- SED (Systemic Exposure Dose) is the estimated amount absorbed through normal cosmetic use
Calculating MoS involves toxicological data and exposure modelling, including body surface area, frequency of use, product type, and skin absorption rates. This approach is based on deterministic exposure modelling, which uses realistic consumer use scenarios rather than worst-case assumptions.
In CPSR evaluations, MoS helps safety assessors decide whether an ingredient is acceptable at a given concentration.
AI formulation tools often lack reliable NOAEL data and expert toxicological judgment, which can lead to unsafe ingredient combinations or incorrect assumptions about product type, making human review essential for accurate safety assessment.
| Ingredient Risk Level | Typical MoS Concern | Outcome Without Correction |
|---|---|---|
| Low-risk humectants (e.g. glycerin) | Rarely an issue | Usually passes |
| Active acids (AHAs/BHAs) | Concentration vs. leave-on exposure | May require reformulation |
| Fragrance allergens | Cumulative exposure across products | Often requires labelling |
| Preservatives | Narrow margin between efficacy and safety | Frequent reformulation trigger |
| UV filters | Strict Annex VI concentration limits | High risk if AI-suggested |
Laboratory Testing and SCCS Guidelines
Before CPSR sign-off, formulas typically need:
- Stability testing: confirming the formula holds its texture and efficacy over shelf life
- Challenge testing: confirming the preservative system controls microbial growth
- Microbial safety testing: checking for harmful bacteria at release
- Packaging compatibility testing: confirming the formula doesn’t react with its container
AI-suggested ingredient combinations, especially new active pairings, often need testing before approval because there is no data on how they behave over time. The SCCS publishes safety opinions that lead to Annex restrictions, and once updated, these rules become legally binding.
AI tools trained on older data may miss these updates, so human review using the latest SCCS guidance is always required.
| Test Type | Purpose | Typically Required For |
|---|---|---|
| Stability testing | Confirms shelf-life performance | All formulas |
| Challenge testing | Confirms preservative efficacy | Water-based/leave-on products |
| Microbial safety testing | Confirms product is free of harmful microbes | All formulas |
| Packaging compatibility | Confirms no reaction with packaging | All formulas, especially actives |
How to Get a CPSR for an AI-Generated Cosmetic Formula in the UK
Step 1: Review the AI Formula
Have a qualified safety assessor review the ingredient list and concentrations against current Annex restrictions and SCCS guidance before any further work begins.
Step 2: Verify Regulatory Compliance
Cross-check every ingredient against the latest Annexes II-VI to confirm that nothing has been restricted since the AI tool’s training data was last updated.
Step 3: Source Documentation and Run Required Testing
Collect SDS, TDS, and CoA documentation from suppliers, then commission stability, challenge, or compatibility testing as needed by the formula.
Step 4: Prepare the PIF and Finalise the CPSR
Compile the Product Information File, have the Safety Assessor complete Parts A and B, calculate the Margin of Safety, and sign off once gaps are resolved.
Step 5: SCPN or CPNP Notification
Once the CPSR is complete, notify the product via the UK’s SCPN portal or the CPNP for EU sales before it goes to market.
| Step | Action | Typical Owner |
|---|---|---|
| 1 | Review the AI formula | Safety Assessor |
| 2 | Verify regulatory compliance | Safety Assessor |
| 3 | Source documentation and run testing | Brand/Lab |
| 4 | Prepare PIF and finalise CPSR | Responsible Person/Assessor |
| 5 | SCPN/CPNP notification | Responsible Person |
CPNP vs SCPN Requirements for AI Cosmetic Products
A product sold in both markets needs separate notifications through the EU’s Cosmetic Products Notification Portal (CPNP) and the UK’s Submit Cosmetic Product Notification (SCPN) system, often with separate Responsible Persons, since the two frameworks operate independently despite their similarities. This applies equally whether the formula came from AI or a human chemist.
| Factor | CPNP (EU) | SCPN (UK) |
|---|---|---|
| Jurisdiction | European Union | United Kingdom |
| Governing regulation | Regulation (EC) No 1223/2009 | UK Cosmetic Regulation 2020 (SI 2020/1466) |
| Responsible Person | EU-based | UK-based |
| Required for AI-formulated products | Yes, same as any formula | Yes, same as any formula |
AI vs Human Formulator: Which Performs Better for Compliance?
AI wins decisively in speed, generating formula concepts in seconds rather than hours of manual work. Human formulators and assessors win on accuracy and regulatory reliability, checking formulas against live, current data rather than a training snapshot.
Formulas that go through human-led correction before submission see a meaningfully higher first-pass CPSR approval rate, mainly because common recurring errors are caught early.
| Factor | AI Formulator | Human Formulator/Assessor |
|---|---|---|
| Speed of formula generation | Very fast | Slower |
| Regulatory accuracy | Inconsistent, data-dependent | High, current |
| Documentation gathering | None | Built into the workflow |
| CPSR sign-off capability | None | Yes |
| Best use case | Early-stage ideation | Compliance-ready development |
AI Cosmetic Formula CPSR Failures & Best Practices (EU & UK 2026)
CPSR failures for AI-assisted formulas come down to a short list:
- Restricted ingredients
- Concentrations above SCCS limits
- Missing supplier documentation
- Unsupported claims like “clinically proven” with no study behind them
Avoiding these comes down to a few habits:
- Treat every AI-generated formula as a draft, not a finished product, and route it through a qualified safety assessor before testing begins
- Cross-check ingredient legality against current Annexes directly rather than trusting an AI tool’s compliance claims
- Build a relationship with a safety assessment provider early in development, rather than treating CPSR as a final checkbox
So, can AI-generated cosmetic formulas actually pass a CPSR? Yes, but the path always runs through a human safety assessor, not around one. The pattern holds across nearly every
AI-assisted formula reviewed: a strong start, a handful of fixable compliance gaps, and approval once those gaps are corrected and documented. Treating AI as the first draft rather than the final answer keeps a launch on schedule, rather than stalling at the CPSR stage.
FAQs
Can AI-generated cosmetic formulas pass a CPSR in the EU?
Yes, but only after a qualified safety assessor reviews the formula
- Corrects any compliance gaps
- Calculates the Margin of Safety
- Signs off under Article 10 of Regulation 1223/2009
Can AI create a legally compliant cosmetic formula?
AI can create a starting formula, but not a legally compliant one on its own. Compliance requires human verification against current Annexes, supporting documentation, and a signed CPSR.
Does AI replace a cosmetic safety assessor?
AI tools can’t legally sign a CPSR, calculate the Margin of Safety, or bear the regulatory liability that a qualified Safety Assessor and Responsible Person carry under EU and UK law.
What documents are needed for CPSR approval?
Typically:
- Safety Data Sheets
- Technical Data Sheets
- Certificates of Analysis
- Allergen and IFRA declarations
- Lab results, such as stability and challenge testing
Are AI skincare formulas legal in the UK?
Once the formula has a complete, signed CPSR, it is registered with SCPN and is held within a complete PIF by a UK-based Responsible Person.
Do SCCS Guidelines apply to AI formulations?
SCCS Notes of Guidance apply to every cosmetic formula sold in the EU or UK, regardless of whether a human, an AI tool, or a combination of both developed it.
