Perspective · Trust

Evidence over slop

AI can now make a face you will trust more than a real one. In beauty, that makes evidence more valuable, not less. Our rule: AI may speed up discovery, but it never gets to decide what we claim.

Conceptual illustration of analytical testing equipment
Conceptual illustration

When seeing is no longer believing

Beauty has always sold with faces. That is now a problem. In a study published in PNAS, people shown real and AI-generated faces picked correctly only 48.2% of the time, about chance; training raised that only to 59%. The same study found AI faces were rated slightly more trustworthy than real ones (Nightingale and Farid, 2022).

Put that next to an industry that sells transformation, and the risk is obvious: synthetic skin, synthetic results and synthetic experts, produced faster than anyone can check them. Industry writers have started calling it the slopification of beauty.

Claims already have rules

None of this is a legal grey area. In the European Union, cosmetic claims must meet six common criteria: legal compliance, truthfulness, evidential support, honesty, fairness and informed decision-making, and claims "should be supported by adequate and verifiable evidence" (European Commission, technical document on cosmetic claims). In India, the advertising code says claims about objectively ascertainable facts "should be capable of substantiation" (ASCI Code, Chapter I).

The rules are being broken anyway. In 2024–25, the Advertising Standards Council of India scrutinised 7,199 advertisements; personal care was the third-largest category of violations, at 5.7%, and 14.8% of influencer violations (ASCI Annual Complaints Report 2024–25).

A model can tell you what is worth testing. Only a test can tell you what is true.

What AI is actually good for

We are not against AI. Used well, it is one of the most useful tools in ingredient discovery. Machine-learning models can screen candidates far faster than a bench can: one recent model predicts anti-ageing peptide activity with strong accuracy on independent test sets (Datta et al., 2025).

But a recent review of AI in cosmetic peptide discovery describes the field as still at an early stage, with limited clinical evidence and a need for better experimental validation (Wang et al., 2026). A prediction narrows the search. It does not prove that a product works on people.

Our rules

  1. AI narrows, tests decide. Models may rank candidates; only lab and human tests support a claim.
  2. No synthetic people. We will not use AI-generated faces, skin, experts or testimonials to represent results.
  3. No synthetic results. Before-and-after images will be real, unretouched and labelled with the conditions they were taken in.
  4. Label what is generated. Illustrative and AI-assisted images on our site are captioned as such, as they are today.
  5. Show the evidence. Every claim will link to what was measured, on whom and against what.

Trust is about to become the scarcest thing in beauty. We think the brands that keep it will be the ones that can show their working.

Questions

Can people tell AI-generated faces from real ones?

Not reliably. In a 2022 PNAS study, people identified AI-generated faces with 48.2% accuracy, about chance, and rated them slightly more trustworthy than real faces.

What evidence do cosmetic claims need?

In the EU, claims must be supported by adequate and verifiable evidence under six common criteria. In India, the ASCI code requires factual claims to be capable of substantiation.

How does Anadoris Biotech use AI?

To help decide which candidates are worth testing. Claims are supported only by lab and human testing, and we do not use AI-generated people or results to represent outcomes.

Sources

  1. Nightingale & Farid (2022). AI-synthesized faces are indistinguishable from real faces and more trustworthy. PNAS 119(8)
  2. European Commission (2017). Technical document on cosmetic claims (Regulation (EU) No 655/2013 common criteria)
  3. ASCI. Code for Self-Regulation of Advertising Content in India, Chapter I
  4. ASCI (2025). Annual Complaints Report 2024–25
  5. Datta et al. (2025). Scientific Reports 15:29036 (machine-learning prediction of anti-ageing peptides)
  6. Wang et al. (2026). Pharmaceuticals 19(10):1547 (AI in cosmetic peptide discovery, review)

Related

A perspective is an argument, not a study. Figures are from the sources listed, which are not Anadoris studies. Anadoris Biotech’s research is at the exploration stage: no Anadoris results or product claims are made here.

Disagree, or want to build this with us?

Start a conversation →