Headlines

​Patients with previously treated advanced KRAS G12C-mutant non-small cell lung cancer (NSCLC) lived significantly longer and saw improved disease control when treated with the next-generation KRAS targeted therapy divarasib, according to researchers from The University of Texas MD Anderson Cancer Center.

An integrated alfalfa tissue-culture system produced faster regeneration and transformation, with a 2.8-fold increase in transgene-positive plants and complete regeneration in all 25 cultivars tested.

A peer-reviewed commentary by Kent Buse and Sarah Hawkes argues that public health institutions should take responsibility for evaluating AI-generated content and treating knowledge as a public good.

The data viewer will provide near-real-time, county-level data sets on the use of cover crops and conservation tillage practices to strengthen decision-making for U.S. agriculture. The post Viridis View to Launch at Sustainable Ag Summit to Track Progress for Regenerative Agriculture appeared first on CropLife.

A UCLA artificial intelligence platform paired with three-dimensional tumor testing identified berberine and enzalutamide as a synergistic combination in four laboratory models of resistant prostate cancer.

A calibration and electromagnetic tracking system achieved sub-millimeter accuracy for real-time augmented reality overlays during 16 robot-assisted pelvic lymph node dissection procedures.

The 2026 World Medical Innovation Forum showcased experimental precision-frozen pig organs, a preclinical pancreatic-cancer immunotherapy, AI-enabled cardiovascular screening and a digitally supported maternal-care model.

A new decomposition study shows Hong Kong-born residents have held a consistent adult life expectancy advantage over Mainland China/Macao/Taiwan-born residents, concentrated in later-life pneumonia and cancer deaths, before the gap narrowed sharply in 2021.

Scientists have uncovered a surprising new clue to why the Mediterranean diet is so good for the heart, brain, and healthy aging. Researchers at USC found that older adults who closely followed the diet had higher levels of two tiny proteins, humanin and SHMOOSE, produced by the mitochondria, the energy-producing structures inside cells. These proteins have been linked to protection against heart disease, Alzheimer's disease, and age-related damage.

A retraining-free generative framework called EndoRare synthesizes diverse images of rare gastrointestinal lesions from a single reference image, improving AI classifiers and novice endoscopists' recognition in an exploratory study.

The Regenerative Medicine Project at our institute is advancing therapeutic strategies for various diseases by introducing a range of modifications to human induced pluripotent stem (iPS) cells.

An AI system called DxO, tested on a new sequential diagnostic benchmark of 629 clinical cases, outperformed participating physicians in accuracy while often reducing estimated testing costs.

An opinion article argues that medical AI should strengthen, rather than displace, bedside observation, clinical reasoning and the physician-patient relationship.

A narrative review maps how machine learning and deep learning now span oral cancer care from smartphone lesion triage and whole-slide computational pathology to prognostic modelling and treatment-response prediction, while warning that most evidence remains retrospective and internally validated.

A simulation-based study reports that a federated Edge-AI framework combining smart neck collars and a hybrid GBoost-LSTM model predicted dairy cattle disease with 95.12 percent accuracy while keeping raw sensor data on participating farms.

Cailey Gleeson / Fierce Healthcare: Baba, which develops AI tools to help Medicare and Medicaid patients connect to care teams and handle advocacy tasks, raised a $40M Series A led by FirstMark  —  Applied artificial intelligence company Baba secured $40 million in series A funding to modernize the Medicare and Medicaid experience.

A laboratory study of primary human lung fibroblasts finds that senescent fibrotic cells from idiopathic pulmonary fibrosis lungs activate antiviral genes after influenza infection but fail to mount coordinated IRF3/STAT1-dependent antiviral control.

A retrospective cohort study of 276 simultaneous pancreas-kidney transplant recipients found that antibiotic decontamination of brain-dead donors was associated with significantly fewer postoperative infections, particularly gram-negative bacteremia and intra-abdominal infections.

Chinese researchers distilled three large neural networks into a lightweight model that identifies Ligustici Rhizoma et Radix decoction pieces and their confusable adulterants from smartphone photos with 99.73 percent accuracy.

A new review of thirty studies finds artificial intelligence showing strong potential across minimally invasive vascular surgery, from risk prediction to autonomous navigation, while warning that most models still lack external validation and real-world testing.

DENVER — Children born to kidney transplant recipients had normal growth and development, but careful planning is needed for transplant recipients who want to become pregnant, according to data presented at ASN Kidney Week.The Transplant Pregnancy Registry International (TPRI), originally named the National Transplantation Pregnancy Registry, was established in 1991 to assess pregnancy outcomes among transplant recipients in the U.S., according to Serban Constantinescu, MD, PhD, coinvestigator of the TPRI research team. Since being renamed in 2016, the TPRI has expanded to include

An international team led by RCSI and The Institute of Cancer Research, London, used an AI tool called iPhenMap to analyse 2,000 colorectal tumour samples, uncovering previously unrecognised vulnerabilities in microsatellite-stable bowel cancer and identifying drug combinations, including the approved targeted drug trametinib, that showed promise in laboratory models.

A qualitative study of ten pharmacists at a Chinese tertiary hospital finds cautious acceptance of AI medication consultation, with pharmacists emphasizing AI's auxiliary value and their own irreplaceable role in personalized care and risk identification.

Peripheral nerves lose much of their ability to repair themselves as bodies age, and a new comparative genomics study points to a single immune-cell pathway as a shared culprit in humans and mice. Writing in the Journal of Translational Medicine,

Scientists discovered that the thick, sticky fluid surrounding glioblastoma tumors can make brain cancer cells smaller, more flexible, and better at invading surrounding tissue. Even more troubling, some of these changes persist after the cells leave the thick fluid, revealing a previously overlooked mechanism that could help this deadly cancer spread.

A new Mamba-based segmentation network combines four-directional state space scanning, spatially guided channel attention, and parameter-free anatomical constraints to outperform U-Net, SwinUNet, and Mamba variants on cardiac, abdominal, and prostate benchmarks while running at 53.8 frames per second.

A systematic review of 11 studies finds that deep learning models using ultrasound show moderate sensitivity and high specificity for predicting axillary lymph node metastasis in breast cancer, but wide prediction intervals and geographic concentration limit immediate clinical use.

After birth, a new biological clock may begin. Continue reading Having babies back-to-back may make the body age faster on Tech Explorist.

The Phase 2 BESTOW clinical trial showed that tegoprubart (a novel anti-CD40L antibody therapy) was better at preserving kidney function than tacrolimus (the current standard of care) in immunosuppressive regimens in adults receiving their first kidney transplant.

Trinity College Dublin researchers found that Inhibitor of Apoptosis Proteins let tumour cells survive chemotherapy, enter premature senescence and recruit wound-healing immunity, while IAP antagonists selectively killed these senescent cells.

A review integrates murine, human, and bovine evidence on epithelial and mesenchymal stem cells in the mammary gland, highlighting mTORC1-dependent metabolic regulation, niche interactions, and experimental therapeutic approaches for bovine mastitis.

A particle swarm optimization strategy that dynamically selects the most informative clients in federated learning matches state-of-the-art methods for medical image classification while keeping standard FedAvg aggregation intact.

Warhound says gene editing and cloning can fix why military dog candidates fail, but questions about genes and oversight remain.

A quasi-experimental study at a Chinese transplant center found that a nurse-led pathway bypassing routine ICU care was associated with earlier feeding, mobilization and shorter hospital stays, though the historical-control design cannot prove causation.

The phase 2 BESTOW clinical trial showed that tegoprubart (a novel anti-CD40L antibody therapy) was better at preserving kidney function than tacrolimus (the current standard of care) in immunosuppressive regimens in adults receiving their first kidney transplant. Kidney graft function stabilized after the first month of treatment and remained higher in tegoprubart-treated patients than in tacrolimus-treated patients at each reported time point, with a statistically significant kidney function benefit at 18 months. A recent analysis of BESTOW data found that tegoprubart also caused greater improvements in patient-reported health-related quality of life compared with tacrolimus in study participants. The findings will be presented at ASN Kidney Week 2026, Oct. 22–25.

A repeated Japanese survey of physicians and the public found that intention to use relatively autonomous medical AI did not change significantly between 2018 and 2025 despite rising self-rated knowledge.

A new narrative review finds that quantifying structural HLA differences between donor and recipient predicts antibody-mediated rejection risk and may guide immunosuppression, though prospective interventional evidence remains limited.

A retrospective French study found vulvovaginal graft-versus-host disease in 25.5% of women after allogeneic stem cell transplantation, often with late onset and severe forms.

The new RFD4-Proteina unified generative model for protein design—created by the Institute for Protein Design (IPD) in collaboration with NVIDIA—uses advanced AI to generate entirely new protein sequences and structures, achieving computational design success rates five to 10 times higher than previous models and generating candidates five to 10 times faster.

Google Health 5.11 is rolling out this week as a rather big update for Fitbit trackers and the Pixel Watch. more…

The start-up is the brainchild of a founder of Anduril, the military tech firm, and Colossal Biosciences, a biotech company known for its de-extinction efforts

How can the identity of pluripotent stem cells be determined more reliably during their differentiation? Researchers from the IUF—Leibniz Research Institute for Environmental Medicine and Heinrich Heine University Duesseldorf have systematically validated marker genes for this purpose and developed a web-based application for the automated classification of early cell fates. The study has been published in Stem Cell Reports.

With the Fitbit Edge announcement, Google Health previewed improved automatic activity tracking, and it’s now beginning to roll out with version 5.11 of the app. more…

Robots the size of a pencil tip could travel through the body to deliver medicine to infections inside a beating heart.

A study tied modest frailty gains in a prehab program to lower waitlist death risk in liver transplant candidates. Here is what caregivers can ask the care team.

A systematic bibliometric analysis of 25 years of Ficus literature shows that only about 11.3 percent of publications address genomic or genetic engineering topics and that CRISPR-related keywords are absent from the field's thematic maps, revealing a translational bottleneck between rapidly expanding genomic resources and functional genome editing in figs.

Researchers propose a testable framework that pairs Traditional Chinese Medicine's preventive-treatment doctrine with digital phenotyping and dual-engine AI analysis for chronic disease control, while stressing the architecture remains clinically unvalidated.

One of the many challenging parts of being a doctor is diagnosing exactly what is wrong with a patient. This is because numerous illnesses often have overlapping symptoms, which can make the diagnostic journey a long one.

Medical AI can improve with new patient data, but every update must be checked first. Patients, equipment and ways of working change, and a model trained on older data can gradually get worse without anyone noticing.

Human cells hold about two meters (6.6 feet) of DNA, folded and compacted to fit inside a nucleus. That packing is essential for life, but it creates a problem for scientists. Some of the genes that control a cell's identity are buried so deep in tightly packed DNA that gene-targeting tools such as CRISPR activation, or CRISPRa, struggle to reach them.

An integrated single-cell atlas of more than 375,000 endothelial cells from 12 human tissues identifies conserved transcriptional hallmarks of vascular aging, including inflammation, reduced angiogenic gene expression and a p21-linked senescence program.

Katherine Davis / Axios: Pathos AI, which uses AI to develop oncology drugs, raised a $237M Series E at a $3B post-money valuation, up from a $1.6B valuation in May 2025  —  Precision oncology biotech Pathos AI said Friday it raised a $237 million Series E round at a $3 billion post-money valuation.

A systematic review of 20 qualitative studies finds that blood cancer patients often experience hematopoietic stem cell transplantation as a constrained choice shaped by family, finances, and fears about identity rather than a free weighing of options.

Microsoft has developed an AI diagnostic tool that solved complex medical cases more than twice as accurately as generalist doctors in a study published in Science. Unlike many medical AI systems, it works through a case step by step, deciding which questions to ask and which tests to order before reaching a diagnosis. The tool, […] This story continues at The Next Web

Ren Xiaoping, who once vowed to perform the world’s first human head transplant, has operated on a patient’s severed spinal cord in a 12-hour landmark operation in Moscow, according to the Chinese surgeon and Russian state sources. Ren said the procedure was carried out at the Sklifosovsky Research Institute for Emergency Medicine in Russia last week. “It was our Guangxi University of Chinese Medicine Second Affiliated Hospital and the Russian team working together,” Ren told the South China...

Take a selfie to see how healthy you are: the idea is enticing and this new research is promising. But so many factors have been ignored The idea that a single selfie could predict how long someone with cancer may live sounds like an incredible breakthrough. AI is getting better at spotting patterns in medical data, but can it be used to reliably predict what will happen to an individual patient? I asked a clinical colleague what she thought about the recent headlines. She found it an interesting development given that doctors have long used facial appearance to assess patient health. For example, she noted that they may write “appears stated age” or “appears older than stated age” as a proxy for general health. Now, researchers from Mass General Brigham, a healthcare institution in Massachusetts, have taken this idea a step further. They have developed an AI model trained on more than 40 million facial images – called FAHR-Face – to assess biological age and predict survival

Unresponsive ICU patients in a new Columbia study were able to raise and lower the frequency of a beeping tone with their minds, via noninvasive EEG electrodes that recorded and translated their brainwaves.

Amazon, Google, Meta, Oracle, and Axon backed electronic prior authorization for self-funded plans. See what could change for workers and families.

arXiv:2610.30470v1 Announce Type: new Abstract: Most voice-cloning systems are built for audio alone: they copy the reference recording's delivery and acoustics, so a voice cloned on a phone sounds like a phone recording whether the speaker is pictured in a studio, on a street, or in a cathedral. Speech for a talking video should sound like the pictured person speaking in the pictured place. We present HeyGen Voice, a scene-aware voice-cloning model that separates who is speaking from how and where: it keeps the speaker's voice from an everyday recording while the target image sets the delivery and acoustics. Prior image-conditioned TTS does half of this job: it either adds scene acoustics without keeping a specific speaker or infers a voice from a face. HeyGen Voice is a multimodal autoregressive transformer conditioned on the image, speaker and scene captions, and the voice recording, trained in five stages that move from multilingual TTS pretraining through simulated and real

arXiv:2610.30229v1 Announce Type: new Abstract: Medical artificial intelligence (AI) models are often shared without their training data, assuming this protects the patients who contributed them. However, a model can reveal whether a patient's records were used in training, a privacy breach known as a membership inference attack. This risk is usually reported as a cohort average, which can appear negligible while individual patients are almost certainly exposed. Identifying potentially vulnerable records before developing a production model could make privacy assessment more practical. We measured each patient's vulnerability to membership inference attack, across 7,789 models trained on six clinical datasets in four modalities, including two adapted pretrained medical models. Training choices changed how many patients had high attack scores, but the patients who ranked highest by risk largely stayed the same across the configurations we tested. Stronger weight decay reduced the

Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’  TechCrunchMicrosoft says its AI tool is better than doctors at diagnosing conditions  The VergeMicrosoft Says Its New AI Is More Than Twice as Good at Diagnosing Diseases as Doctors  GizmodoMicrosoft Diagnostic AI Outperforms Generalist Physicians in Science Study  Unite.AICan AI help doctors avoid missed diagnoses? A new study suggests yes  TAPinto

Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’  TechCrunchMicrosoft says its AI tool is better than doctors at diagnosing conditions  The VergeMicrosoft Says Its New AI Is More Than Twice as Good at Diagnosing Diseases as Doctors  GizmodoMicrosoft Diagnostic AI Outperforms Generalist Physicians in Science Study  Unite.AICan AI help doctors avoid missed diagnoses? A new study suggests yes  TAPinto

The presence of non-native trees is often seen as an obstacle to the recovery of tropical forests regrowing in previously deforested areas. To better understand the behavior of these plants over time in both humid and dry forests, researchers from approximately 60 institutions analyzed data from over 1,500 areas of forest regeneration across the Americas. The results, published in the journal Nature Plants, indicate that the initial dominance of exotic species is temporary. They multiply easily during the first 10 to 20 years, taking advantage of abundant sunlight in open areas. However, they decline as native trees grow and shade the forest.

Natural disasters. Hikes. Long, dark nights alone. For many people, dogs offer a sense of safety and comfort in challenging or uncertain times. Cancer researchers at a new lab at Purdue University are partnering with dogs to make cancer less frightening—and less deadly—to some of humanity's very best friends using a technique called tumor ablation.

Meet Turritopsis dohrnii, a tiny jellyfish that can revert to an earlier stage of its life cycle when times are tough, offering insights into how to reverse aging.

The pancreas is a complex organ comprising multiple cell types that regulate blood glucose levels through hormones, such as insulin, and support digestion by producing digestive enzymes.

UCLA investigators have developed a new artificial intelligence-based platform that could help scientists identify promising drug combinations to overcome treatment resistance in cancer.

Huntington's disease is caused by repetitive "stutters" in DNA—specifically repeated CAG sequences—that expand abnormally over time, leading to toxic protein accumulation that contributes to the progression of the disease. There is currently no cure for this rare and fatal genetic disease.

A groundbreaking immunotherapy treatment for rectal cancer, developed through pioneering research at Memorial Sloan Kettering Cancer Center (MSK) with critical early-stage funding from Swim Across America, has received accelerated approval from the U.S. Food and Drug Administration (FDA).

Scientists from Nanyang Technological University, Singapore (NTU Singapore) have developed a new class of light-activated compounds that could lead to safer and more versatile applications in 3D printing, cosmetics and medicine.

Microsoft’s new AI diagnostic system beat generalist doctors on hard cases, according to a study in Science. But most of the improvement seems to come from the AI models underneath, not from Microsoft’s system itself.

Microsoft AI chief Mustafa Suleyman. | Getty Images For years, AI leaders have touted healthcare as one of the clearest examples of where their technology will improve lives. Now "it is finally here," Microsoft AI CEO Mustafa Suleyman tells The Verge. The company claims to have developed an AI-powered tool it says can diagnose conditions up to four times more accurately than human doctors. The findings, published Thursday in one of the world's most prestigious journals, Science, are based on tests pitting Microsoft's AI system against human physicians on a series of scenarios drawn from medical case reports and emergency department data. Microsoft compiled the cases into its own benchmark, S … Read the full story at The Verge.

Conventional wisdom suggests that the long hours and pressures of entrepreneurship can take a toll on health. But new research by Cristiano Bellavitis, associate professor in the School of Business at Stevens Institute of Technology, and his collaborators finds that entrepreneurs tend to live longer than people who spend their careers working for others.

Elon Musk threatens legal action over Alex Gibney's documentary, which uses an AI-generated likeness and faces disputes over Starlink and the 2024 US election.

Unresponsive ICU patients in a new Columbia study were able to raise and lower the frequency of a beeping tone with their minds, via noninvasive EEG electrodes that recorded and translated their brain waves.

AI adoption in medicine is accelerating: More than 12,000 primary care clinicians have enrolled in Canada Health Infoway's national AI scribe program, about 5,000 Ontario physicians use AI scribes and roughly one in five Canadian physicians now use AI tools. Almost 60% report a cut in administrative time.

Rebecca Torrence / Bloomberg: Warp Sciences, founded by ex-OpenAI product chief Kevin Weil to speed up AI drug development, emerges from stealth with $215M at an $865M post-money valuation  —  Kevin Weil, the former OpenAI executive who led the ChatGPT maker's AI-for-science efforts, is launching a new startup aimed …

As we get closer to autonomous, AI-driven medical care becoming a reality, regulators are racing to catch up. The FDA in August released a discussion paper with ideas about how ...

Human neural stem cells that help form the brain and spinal cord are highly susceptible to SARS-CoV-2, including omicron variants BA.1 and BA.2, according to new research from the University of California, Riverside. The findings raise concerns about the potential effects of maternal COVID-19 infection on early nervous system development and underscore the importance of monitoring children exposed to the virus before birth.

Researchers at the Johns Hopkins Sidney Kimmel Comprehensive Cancer Center and Johns Hopkins Children's Center report that bone marrow transplant using half-matched family donors achieved a 96% survival rate among children, adolescent, and young adult patients with severe aplastic anemia, a rare, life-threatening disorder in which the bone marrow fails to produce enough healthy blood cells.

An in-vitro study found that resveratrol glucoside reduced senescence-associated cellular changes by altering mitochondrial metabolism and chromatin regulation while showing less cytotoxicity than resveratrol.

Researchers analyzing nearly 150,000 Reddit posts uncovered unexpected symptom reports linked by users to unapproved versions of retatrutide, including increased appetite, sleep disturbances, and elevated heart rates. The findings differ from clinical trial results and highlight how AI could help identify potential drug safety signals that warrant further investigation.

How much evidence should the FDA require before it authorizes a generative AI-enabled medical device? The question was a point of contention among healthcare lobbying organizations and tech companies, who weighed in as the FDA ...

Potential and confirmed victims of human trafficking from Ukraine have been identified across Europe, Dmitry Polyansky stressed

A controlled study of 70 male rats found that stem cells led early wound contraction, while chitosan achieved the strongest contraction and marker results by day 14.

A review finds that electroactive tissue-regeneration scaffolds must be evaluated through linked electrical, mechanical, chemical and biological properties to address translational limitations.

A theory-based synthesis proposes five user and workforce functions, organizational controls and multilevel safeguards for safer decisions from conversational generative-AI health information.

A review identifies crosslinking as a multidimensional design variable influencing mechanics, transport, degradation and cell interactions in preclinical biopolymer biomaterials.

Researchers have outlined a plan for slowing aging by targeting one of the body's most important cellular communication systems. In a new paper published in Frontiers in Science, scientists describe how novel therapies that modulate GPCR signaling could one day help delay multiple age-related diseases at once—potentially extending healthy years of life.

A nationwide Danish twin study estimated that additive genetic effects explain about half of the variation in liability to clinically diagnosed rotator cuff tears.

arXiv:2610.30119v1 Announce Type: cross Abstract: We prove a self-normalized Bernstein inequality for bounded functions of a regenerative process with independent, identically distributed marked cycles of finite mean length. The bound uses observed cycle lengths and retains explicit terms for truncation and the cycle ending after the observation horizon. Selecting a tuning parameter from a finite deterministic collection gives a version of Bernstein inequality where all terms are empirical but one, related to the tails of the regeneration times. We then study this term in many regimes: it turns out that it is of second-order in many cases of interest. This is important for applications in machine learning: we prove generalization error bounds for classification with non-i.i.d., regenerative data. These bounds can be used to define learning objectives: by minimizing the first-order term only in the bound, we define agnostic classifiers that still satisfy oracle inequalities. We also

arXiv:2610.30119v1 Announce Type: cross Abstract: We prove a self-normalized Bernstein inequality for bounded functions of a regenerative process with independent, identically distributed marked cycles of finite mean length. The bound uses observed cycle lengths and retains explicit terms for truncation and the cycle ending after the observation horizon. Selecting a tuning parameter from a finite deterministic collection gives a version of Bernstein inequality where all terms are empirical but one, related to the tails of the regeneration times. We then study this term in many regimes: it turns out that it is of second-order in many cases of interest. This is important for applications in machine learning: we prove generalization error bounds for classification with non-i.i.d., regenerative data. These bounds can be used to define learning objectives: by minimizing the first-order term only in the bound, we define agnostic classifiers that still satisfy oracle inequalities. We also

arXiv:2610.30119v1 Announce Type: new Abstract: We prove a self-normalized Bernstein inequality for bounded functions of a regenerative process with independent, identically distributed marked cycles of finite mean length. The bound uses observed cycle lengths and retains explicit terms for truncation and the cycle ending after the observation horizon. Selecting a tuning parameter from a finite deterministic collection gives a version of Bernstein inequality where all terms are empirical but one, related to the tails of the regeneration times. We then study this term in many regimes: it turns out that it is of second-order in many cases of interest. This is important for applications in machine learning: we prove generalization error bounds for classification with non-i.i.d., regenerative data. These bounds can be used to define learning objectives: by minimizing the first-order term only in the bound, we define agnostic classifiers that still satisfy oracle inequalities. We also show

arXiv:2610.29351v1 Announce Type: new Abstract: Zero-shot cross-lingual voice cloning often carries the reference speaker's accent into the target language. We propose closed-loop dual-reference guidance, which pairs the original reference with a reusable native anchor constructed by voice conversion (VC). The references initialize separate language-model (LM) states that guide a shared acoustic trajectory; each completed patch then updates both states, keeping their generated histories aligned with the emitted speech. Anchor construction runs once per speaker, and online synthesis uses a frozen TTS model without per-utterance VC. On 20 held-out French speakers, the fixed setting $\lambda^{\!*}=0.33$ reduces residual French LID mass by 73\% versus direct cloning, at an ECAPA similarity cost of only $0.042$; relative to anchor-only cloning, it recovers $0.155$. The method achieves the lowest mean WER ($0.009$) and highest mean GOP ($4.750$) among the held-out test arms. In blinded

arXiv:2610.29013v1 Announce Type: new Abstract: Speculative decoding accelerates large language models by verifying drafted tokens in parallel. New draft-training objectives are usually credited with the acceptance they add over cross-entropy on externally written responses, yet many pipelines let the target rewrite them. We ask how much of that gain survives. With a vision-language target and a block drafter, we train three objectives on four versions of one corpus: cross-entropy on corpus tokens, token-level distillation from the target, and distillation gated to positions that would survive verification. Distillation and the gate add large gains on external responses, but after greedy regeneration, distillation adds a smaller gain and the gate little or none. The shrinkage persists across data orders, position weights, verifiers, decoding temperatures and, for distillation, initializations and learning rates. Most of it remains at matched supervised length and at each objective's