What should I make of an AI drug headline?
AI can change how researchers look for a drug candidate. It does not change the need to prove that candidate works, in the people it is meant to help. Evidence
What is known
- Reasonably supported
FDA and EMA jointly published ten guiding principles for AI use in drug development in January 2026. The first principle: AI applications should be human-centered by design, with a person still required to check the work and protect patient safety.
What is uncertain
- Depends on you
A computer-assisted candidate is still only a candidate. Safety, benefit, and who it may help still need to be studied and reviewed before it becomes a real care option. A headline often blurs discovery, testing, and approval into one story.
Questions for your dermatologist
Was this candidate actually studied for psoriasis, or is the story borrowing hope from a different condition?
Saving keeps this on your device and needs JavaScript, which is off in this browser.Is there a dated original announcement or study behind this headline, not a repost of one?
Saving keeps this on your device and needs JavaScript, which is off in this browser.
Two real examples, and neither is a psoriasis treatment
Two real AI-assisted candidates show where this research actually stands. Rentosertib is in Phase III testing for a lung disease. BEN-2293 met its safety goal but not its efficacy goal, in eczema. Neither one is being studied in psoriasis. Evidence Evidence
What is known
- Limited evidence
Insilico Medicine says AI helped find rentosertib's target and design its shape. The company began a roughly 320-patient study in July 2026, for a lung-scarring disease. BenevolentAI says its AI-assisted skin cream BEN-2293 passed its safety check in a 91-patient study of eczema. That study did not show a clear benefit on the planned itch and skin scores, across the whole group.
What is uncertain
- Depends on you
Both results come from the company that developed the candidate, not from an independent or peer-reviewed report checked here. A safety pass is not a benefit finding, and starting Phase III testing is not an approval. As of this check, no source found here studies either candidate in psoriasis.
Questions for your dermatologist
Does either of these change my plan today, or should we keep going the way we discussed?
Saving keeps this on your device and needs JavaScript, which is off in this browser.If a psoriasis-specific AI-assisted study ever opens, how would I find out?
Saving keeps this on your device and needs JavaScript, which is off in this browser.
Why show a candidate that missed its goal?
Honest tracking includes a candidate that stalls or misses its goal, not only the ones that succeed. Otherwise a pipeline page reads like advertising. Evidence
What is known
- Reasonably supported
ClinicalTrials.gov holds study records that sponsors submit themselves, whatever the eventual result. A record there is a place a study is listed, not an approval and not a government endorsement of the treatment.
What is uncertain
- Depends on you
A listing does not say a candidate is safe, effective, or close to reaching anyone. It does not say whether a psoriasis-specific AI-assisted candidate exists in the registry today.
Questions for your dermatologist
Is the thing I read about approved, or still just a study record?
Saving keeps this on your device and needs JavaScript, which is off in this browser.What source would you trust for an update on this later?
Saving keeps this on your device and needs JavaScript, which is off in this browser.
Evidence behind this page
Sources
Each evidence badge opens the source and its limits. The full list stays available here.
- U.S. National Library of MedicineRegulatory · Regulatory / guideline, tier 1Independent source
- Published
- SteadySkin last checked
What this source can and cannot tell you
What this source supports
Supports that ClinicalTrials.gov is a U.S. government database whose sponsor-submitted study records can include status, eligibility, locations and contacts.
What it does not support
A listing is not government approval, scientific validation, proof of benefit, a completeness guarantee or a determination that a reader qualifies.
Claim-specific review for this source is still in progress. Only the source-level evidence and limits are shown here.
- FDA Center for Drug Evaluation and Research, FDA Center for Biologics Evaluation and Research, and the European Medicines AgencyRegulatory · Regulatory / guideline, tier 1Independent source
- Published
- SteadySkin last checked
What this source can and cannot tell you
What this source supports
Supports that FDA and EMA jointly published ten guiding principles for AI use in drug development on January 14, 2026. Supports that the first principle calls AI applications human-centered by design. It requires human involvement to protect patient safety and the reliability of what AI produces.
What it does not support
It is a general policy framework. It is not a decision about any one candidate drug. It does not evaluate rentosertib, BEN-2293, or any drug studied for psoriasis. It does not say AI has found a psoriasis treatment.
Claim-specific review for this source is still in progress. Only the source-level evidence and limits are shown here.
- Insilico MedicineManufacturer document · Manufacturer, tier 4Relevant relationship disclosed
- Published
- SteadySkin last checked
What this source can and cannot tell you
Why the independence label says this: This is the drug's own developer announcing its own candidate's trial progress, not an independent or peer-reviewed report.
What this source supports
Supports that Insilico Medicine says AI helped find rentosertib's target and design its shape. Supports the company's own report: lung function rose more on the drug than on placebo at 12 weeks, in an earlier study. Supports that the company started a roughly 320-patient study on July 7, 2026, across 47 sites in China. Supports that this new study is for a lung-scarring disease, not psoriasis.
What it does not support
It is the sponsor's own announcement about its own drug. It is not an outside or peer-reviewed check of the earlier result. Starting a new study is not an approval. It is not proof the drug works or is safe. No record found here studies rentosertib in psoriasis.
Claim-specific review for this source is still in progress. Only the source-level evidence and limits are shown here.
- BenevolentAIManufacturer document · Manufacturer, tier 4Relevant relationship disclosed
- Published
- SteadySkin last checked
What this source can and cannot tell you
Why the independence label says this: This is the drug's own developer announcing its own candidate's trial results, not an independent or peer-reviewed report.
What this source supports
Supports that BenevolentAI calls BEN-2293 a topical, AI-assisted candidate. Supports the company's own report: a 91-patient study found it safe and well tolerated. Supports that the study did not show a clear benefit on itch or skin inflammation across the whole group. Supports that a later look at patients with more skin involved hinted at a possible benefit worth further study. Studied for mild-to-moderate eczema, not psoriasis.
What it does not support
It is the sponsor's own announcement, not an outside or peer-reviewed check. A safety pass is not a benefit finding. The study did not meet its benefit goal in the whole group. A later, smaller finding is not a confirmed one. No record found here studies BEN-2293 in psoriasis.
Claim-specific review for this source is still in progress. Only the source-level evidence and limits are shown here.