New Autism Treatment Research in 2026: What Parents Should Know

The short answer

That answer may sound less exciting than the headlines. It is also more useful. A family should not have to discover—after paying a clinic, importing a product or raising a child's hopes—that “promising” meant one small study, one secondary outcome or a company press release.

Autism is a lifelong neurodevelopmental difference, not an illness that needs curing. Autistic children may still need help with communication, sleep, anxiety, pain, epilepsy, ADHD, eating, distress or daily living. Good support aims to improve wellbeing, safety, autonomy and participation—not to make a child appear less autistic.

What does “new autism treatment” actually mean?

The phrase can describe very different things:

• A new laboratory idea: interesting, but not tested in people.
• An early clinical trial: a safety check or first signal, not proof.
• A positive Phase 2 trial: a reason to run a larger confirmatory trial.
• A successful Phase 3 trial: stronger evidence, but still not automatically a UK licence or NHS recommendation.
• An available treatment: something a properly regulated service can offer for a clear indication, with known monitoring and realistic goals.

Those steps are not interchangeable. A product can be “in Phase 3” and still fail. A clinic can sell something privately even when evidence is weak. A medicine can help a coexisting condition without treating autism itself.

The 2026 evidence at a glance

Why the main outcome matters: researchers decide a trial's primary outcome before seeing the results. This reduces the risk of testing many measures, then highlighting whichever one happens to look positive. If the primary outcome is negative, secondary findings do not establish the trial's main claim. They may still help researchers decide what to test next.

AST-001: early promise, then a Phase 3 miss

AST-001 is a proprietary syrup formulation based on L-serine. A South Korean Phase 2 trial involved 151 autistic children aged 2 to 11. After 12 weeks, the high-dose group improved by an adjusted 1.43 points more than placebo on the trial's main adaptive-behaviour scale. The confidence interval was narrow but used a 90% level, and other familiar measures, including the Social Responsiveness Scale and Aberrant Behavior Checklist, did not show a significant difference.

That was a preliminary signal—not a large change, and not proof that parents should buy L-serine. The most important later update reported by the sponsor came in August 2026: its Phase 3 study did not meet the pre-specified 12-week primary endpoint. Detailed Phase 3 results were not yet peer reviewed or posted on the registry when this page was checked.

The sponsor plans another confirmatory study. That is legitimate research. It also means the honest answer today is “not established”. Read our full AST-001 autism evidence guide.

Arbaclofen: secondary findings after a negative primary result

Arbaclofen acts on the brain's GABA-B system. A 2026 European Phase 2 trial randomised 123 autistic children and teenagers with fluent or complex speech. It ran at seven sites in Spain, France and the UK for 16 weeks.

The main Vineland socialisation outcome did not differ significantly from placebo: the adjusted difference was 1.30 points, with a 95% confidence interval from −2.6 to 5.1. The key clinician severity measure was also negative. Several parent and teacher questionnaires favoured arbaclofen, but the paper explains that these multiple comparisons were not corrected. That raises the chance of a positive-looking result appearing by accident.

Sleep-related problems were reported in 57.6% of participants taking arbaclofen and 34.9% taking placebo. The results justify more study, not routine prescribing. Arbaclofen is also not the same product as ordinary baclofen. See our plain-English analysis of the 2026 arbaclofen trial.

NTI164: a positive result that needs careful reading

NTI164 is a proprietary full-spectrum medicinal-cannabis extract with very low THC. In August 2026, an Australian team published an eight-week placebo-controlled study involving children and teenagers with higher support needs.

The main clinician-rated severity score favoured NTI164 by 1.65 points. Some adaptive-function and social-responsiveness measures also favoured it. That is a positive primary result in this one study; it does not yet establish meaningful or lasting benefit.

But the blinded comparison analysed only 54 children who completed treatment, at one specialist clinic, over eight weeks. Seven of the 61 enrolled participants did not complete that phase. Many outcomes were tested, long-term safety remains unknown, and the study had company funding and author links to the product's developers. Some Social Responsiveness Scale outcomes—including the total score—were not statistically significant.

Most importantly for UK parents, NTI164 is not an over-the-counter CBD oil and is not available outside Australia. A larger Phase 3 trial is planned. Read our detailed NTI164 and autism evidence check or our wider UK guide to CBD and medical cannabis.

Theta-burst brain stimulation: two different short-term signals

Two sham-controlled trials published in 2026 tested accelerated theta-burst stimulation, a form of transcranial magnetic stimulation (TMS). Both were conducted in China. They used different stimulation patterns, brain targets, schedules, age groups and main outcomes, so the second study did not replicate the first.

The BMJ trial randomised 200 children aged 4 to 10 to active or sham accelerated continuous theta-burst stimulation. The schedule was intense: ten sessions a day for five days, or 50 sessions. Parent-rated social-responsiveness scores favoured active treatment immediately and one month later. A storytelling task also favoured treatment among the 119 children able to complete it. This was not an overall language result: 79 children could not complete that task, and general language, receptive vocabulary and Vineland communication did not differ from sham. Adaptive behaviour did not differ either.

Follow-up stopped after one month. Adverse events were reported by 54.5% of the active group and 29.3% of the sham group; restlessness and scalp discomfort were most common. One moderate arm spasm resolved within minutes, and no seizures were reported. Children with an IQ below 50 and children unable to tolerate a practice session were not enrolled.

A separate Molecular Psychiatry trial randomised 67 autistic people aged 6 to 30 to active or sham accelerated intermittent theta-burst stimulation; 59 were in the main analysis. Most had an IQ below 70. This was a much longer programme: three sessions a day on 60 treatment days, or 180 sessions over 12 weeks, plus two individual behavioural sessions each treatment day in both groups. At week 12, 21 of 38 people in the active group and 6 of 21 in the sham group met the study's clinician-rated social-affect response rule: 55% compared with 29% (p=.049). There was no group difference at week eight.

Overall language, parent-rated social communication, repetitive behaviour and emotional or behavioural measures did not differ in the second trial. A language finding appeared only in a subgroup described as having profound autism after the researchers looked separately at the data. That post-hoc result is a lead for future research, not evidence that the treatment improves language. There was no follow-up after the 12-week course. One participant was admitted to hospital with pneumonia, judged unrelated to treatment; local pain was reported in 5% of the active group.

The second trial also started before it was publicly registered. Its registry history shows that the planned sample and main outcome changed during recruitment, and the response definition changed again after recruitment was reported to have ended. This does not prove the result is wrong, but it makes a borderline p=.049 result less secure because the final outcome rule was not publicly fixed before enrolment.

Neither trial had a UK site. Neither protocol is recommended in NICE autism guidance or part of a routine NHS autism pathway. These studies are stronger evidence than an uncontrolled clinic testimonial, but they are not evidence that a UK family should buy TMS or a home electrical-stimulation device. Our updated neurofeedback and brain-stimulation guide explains the techniques, safety limits and UK position.

Genetic and metabolic assessment: diagnosis for selected children, not an autism cure

The phrase “precision treatment” can create the wrong impression. There is no DNA test that diagnoses autism or predicts which general autism treatment will work. Genetic testing can sometimes identify a separate rare condition when an autistic child also has features such as more significant unexplained developmental delay or intellectual disability, seizures, congenital differences, unusual growth, distinctive physical features or loss of skills.

In NHS England, isolated autism—or autism with only mild intellectual disability—is not currently a genomic-testing indication. A paediatric, neurology, genetics or metabolic team looks at the whole clinical pattern and decides whether a selective test is appropriate. Scotland, Wales and Northern Ireland have their own pathways.

A clear rare-disease diagnosis may change seizure care, medical monitoring or, occasionally, lead to a condition-specific diet or medicine. That treats the underlying disorder or a complication; it does not remove autism. Metabolic testing is also guided by clinical and biochemical clues rather than offered as a blanket panel to every autistic child.

If a child loses previously acquired skills, seek prompt medical assessment rather than buying a commercial “autism genetics” panel. Our full autism DNA test and genetic-testing guide explains the current NHS England criteria, possible results, urgent red flags and commercial-test limitations.

L1-79: a negative main outcome beside positive secondary signals

L1-79 is an experimental oral medicine that reduces production of catecholamines, including dopamine, noradrenaline and adrenaline. The theory that this will improve autism-related social communication remains unproven.

The only full peer-reviewed human results paper identified was an uncontrolled case series of eight people. An earlier 42-person Phase 2 trial was designed mainly for four-week safety and was too small to establish benefit. In a later 58-person US Phase 2 trial, carryover and treatment-sequence problems meant the planned crossover analysis could not be used. The registered Vineland responder outcome did not significantly differ from placebo. A continuous Vineland secondary score and several other measures favoured L1-79 in a smaller first-period analysis.

Those later findings were available only as a conference abstract. Numerical results and adverse-event tables were not posted on ClinicalTrials.gov, and no full peer-reviewed results paper or Phase 3 trial was identified when checked. L1-79 remains investigational, has no established UK autism use and has not been properly tested in controlled trials in children under 12. Read the full L1-79 autism evidence check.

Bumetanide: possible responder profiles are not a clinical test

Bumetanide's two large Phase 3 trials involved 422 children and found no benefit over placebo on their main CARS-2 outcome or key secondary outcomes. A 2026 paper used machine learning to search those same datasets for smaller symptom-defined groups that might have responded.

Three proposed profiles produced a repeatable signal on the parent-rated SRS-2 secondary outcome. But no subgroup reproduced a benefit on CARS-2, the Phase 3 primary outcome. The profiles were developed and checked within the same two existing trials and have not been confirmed in a new, pre-registered study. They are research leads—not a questionnaire, biomarker or reason to prescribe bumetanide to a particular child. Our updated bumetanide and autism guide explains the analysis, electrolyte risks and UK monitoring context.

Two more 2026 results parents may see in headlines

Pimavanserin

Pimavanserin was tested for irritability in autistic children and teenagers, not as a general treatment for autism. Neither tested dose beat placebo on the main irritability measure after six weeks, and secondary outcomes did not show a significant advantage. A negative trial is still useful: it helps prevent ineffective treatment from being adopted on hope alone.

ML-004

ML-004 is an experimental formulation of zolmitriptan. The company reported that a Phase 2 study in autistic adolescents and adults missed its primary social-communication outcome. A very small pre-specified subgroup of irritable adolescents produced nominally positive signals. Those results came from 20 people in the subgroup and had not been peer reviewed when checked. They should be treated as a clue for future research, not evidence of benefit.

Three other medicine stories worth knowing

Leucovorin: a narrow rare-disease approval, not an autism approval

Leucovorin continues to appear in autism-treatment discussions. A small 2018 study in 48 autistic children with language difficulties reported a verbal-communication signal. But a larger 2024 study was retracted in January 2026, and no large multicentre Phase 3 trial has established routine benefit or a validated responder test.

The US FDA approved leucovorin in March 2026 for genetically confirmed FOLR1-related cerebral folate transport deficiency, a rare disorder—not for autism generally. Commercial folate-receptor-antibody or MTHFR tests do not show that an autistic child will respond. Read the full leucovorin (folinic acid) and autism UK evidence guide.

Memantine: a small new signal beside an older negative trial

A 2025 single-centre trial in autistic 8-to-17-year-olds without intellectual disability reported more composite “responders” with memantine than placebo. Only 35 participants had post-baseline data, the confidence interval was wide, and continuous social-responsiveness and global-functioning scores did not significantly differ. An earlier, larger study of 121 children was negative. This is not enough to make memantine an autism treatment, but it may inform better-targeted research.

CM-AT: selected for a future NICE appraisal, not recommended

CM-AT is an investigational enzyme product studied for irritability in young autistic children. An industry-funded Phase 3 study of 190 children reported a modest advantage on the Aberrant Behavior Checklist irritability scale. NICE selected the product for a possible future technology appraisal, but the project remains “awaiting development”. Selection means NICE may assess it if regulatory development proceeds; it is not approval, guidance or NHS availability.

What does the research say about speech, language and communication?

None of the medicines or devices in this guide is established as a way to develop spoken language. Communication is broader than speech and should not be judged by speech alone. Trial scales also do not always measure the same thing.

The small 2018 leucovorin trial reported an average verbal-communication difference in children who already met its language-study criteria, but it has not been confirmed in a large Phase 3 trial or turned into a validated way to identify responders. AST-001 had a positive communication-domain secondary result at high dose in Phase 2, then missed its Phase 3 primary outcome. In the BMJ theta-burst trial, a storytelling task favoured treatment only among the 119 children able to complete it; general language and receptive vocabulary did not differ. In the separate Molecular Psychiatry trial, overall language did not differ, and the positive language finding came only from a post-hoc subgroup. Neither trial establishes an overall language benefit.

Communication support does not need to wait for a drug. A speech and language therapist can assess understanding, speech, interaction and access to augmentative and alternative communication (AAC). AAC can include signs, pictures, communication books or electronic devices; it gives a child a way to communicate now and should not be withheld out of fear that it will prevent speech. See what children's speech and language therapy does and the Kent speech and language therapy directory.

So what is actually available in the UK?

None of the experimental candidates above is a routine NHS or NICE-recommended treatment for the core features of autism. Neither theta-burst trial had a UK site, and NICE autism guidance does not recommend TMS or theta-burst stimulation. NICE also says antipsychotics, antidepressants and anticonvulsants should not be used to manage core autism features in children and young people.

Selective genetic or metabolic assessment is different: it may be available through an NHS pathway when a child's additional clinical features meet current criteria. Its purpose is to investigate a possible rare condition, not to provide a routine autism treatment.

That does not mean “nothing can help”. Current care should start with the individual problem:

• Communication support can include speech and language therapy and AAC.
• An occupational therapist may help identify sensory and daily-living barriers and suggest practical or environmental changes; benefit varies by child and goal.
• Play-based social-communication support involving parents, carers or teachers can be considered.
• Anxiety may respond to properly adapted psychological support.
• Sleep, ADHD, epilepsy, pain, constipation and other coexisting conditions deserve ordinary clinical assessment and treatment.
• Severe distress or behaviour that risks harm needs a careful search for pain, communication barriers, demands, sensory overload and other triggers—not an automatic “autism medicine”.

Our first 100 days after diagnosis guide is a practical place to start. For trials and headlines, use the SENDPath five-question evidence checklist.

Five questions to ask about the next treatment headline

1. Did the trial meet its pre-declared primary outcome? If not, treat subgroup and secondary findings as exploratory.
2. How many children completed the blinded comparison? Fifty children at one clinic tells us much less than a large independent multicentre study.
3. Was the change meaningful in daily life? A statistically significant questionnaire score is not automatically a noticeable improvement in communication, safety or independence.
4. Who funded the study and who holds the product rights? Industry funding does not make a result false, but independent replication matters.
5. What is the exact product's UK status? Ask separately whether it is authorised for that use, recommended by NICE, routinely offered by the NHS, or sold only privately. These are not equivalent.

The bottom line

In 2026, NTI164 and two different accelerated theta-burst protocols produced findings worth studying further. The TBS trials do not confirm each other, neither found an overall language benefit and neither answers the question of lasting benefit. L1-79's registered main responder outcome was negative despite some secondary signals. Bumetanide's proposed responder profiles remain unvalidated after negative Phase 3 trials. Arbaclofen produced uncertain secondary signals after missing its main outcome. AST-001's early promise did not carry through to its Phase 3 primary result. Pimavanserin and ML-004 also missed their main tests.

Genetic or metabolic assessment can be clinically useful for a selected child with additional signs of a rare disorder. It should be described as careful diagnosis that may occasionally change condition-specific care—not as a new treatment for autism.

This is not a reason to abandon research. It is a reason to describe research accurately. Families deserve curiosity without hype, access without exploitation and support that is useful now.

---

Sources and further reading

• NICE CG170 — autism in under-19s: support and management, recommendations
• Kim et al. — AST-001 Phase 2 randomised clinical trial
• ClinicalTrials.gov — AST-001 Phase 3 study record
• Astrogen — sponsor's August 2026 Phase 3 statement (Korean)
• Parellada et al. — 2026 arbaclofen Phase 2 trial
• Keating et al. — 2026 NTI164 Harmony trial
• Tan et al. — 2026 accelerated theta-burst stimulation trial
• ClinicalTrials.gov NCT05927792 — registry record for the five-day a-cTBS trial
• Liu et al. — 2026 personalised frontoparietal accelerated intermittent theta-burst stimulation trial
• ClinicalTrials.gov NCT05890846 — registry record and history for the 12-week a-iTBS trial
• NHS Genomics Education Programme — when genomic testing may be considered for an autistic child
• NICE CG128 — selective medical investigations during autism assessment
• Rothman et al. — eight-person uncontrolled L1-79 case series
• ClinicalTrials.gov NCT02947048 — early four-week L1-79 Phase 2 safety study
• ClinicalTrials.gov NCT05067582 — later L1-79 Phase 2 study, no results posted when checked
• Megerian et al. — L1-79 Phase 2 conference abstract (abstract 343)
• Fuentes et al. — two negative bumetanide Phase 3 trials
• Rabiei et al. — 2026 retrospective bumetanide subgroup analysis
• Pimavanserin Phase 2 trial for irritability in autistic children and adolescents
• MapLight — company-reported ML-004 Phase 2 topline results
• Joshi et al. — small 2025 memantine randomised trial
• CM-AT Phase 3 trial for irritability in young autistic children
• NICE — CM-AT technology-appraisal project, awaiting development
• White et al. — systematic review of AAC and speech production in autistic people
• Royal College of Speech and Language Therapists — AAC guidance

Disclaimer: Information only, not medical advice. Evidence and source status were checked on 30 August 2026. Research can change; speak to an appropriately qualified clinician before starting, stopping or changing a medicine, supplement or treatment.