Friday Jul 31
HEALTHOther

Pharma Spends $7B, Ships Zero AI Drugs

31JUL
$7B0 DRUGS

Every major drugmaker is racing into AI drug discovery right now. Not one has an approved AI-designed medicine to show for it. The bill for that race already tops $7 billion.

Insilico Medicine alone signed pharma deals with Servier, Eli Lilly, SK Biopharma, and Takeda since January. Those four partnerships make up most of the total.

Insilico's rentosertib, the furthest along, only reached a Phase IIa efficacy signal. Nothing has cleared Phase III, the stage that actually decides approval. Early-stage success rates still match the industry's historic average.

The technology isn't failing, human trials are just slow and unforgiving. Investors are pricing this category like that problem is already solved.

full brief & sources

⚡ Why this matters

  • This is the clearest gap yet between AI drug discovery's funding and its actual output.
  • It's a warning for anyone valuing an AI biotech on pipeline size instead of trial results.
  • The bottleneck was never finding candidates, it's proving they work in humans.

🔍 What happened

  • Since January 2026, pharma companies committed over $7 billion to AI drug discovery deals with Insilico Medicine alone.
  • Partners include Servier in oncology, Eli Lilly in oral therapeutics, SK Biopharmaceuticals in neuroimmune disease, and Takeda.
  • Zero AI-discovered drugs are approved for patients anywhere, as of July 2026.
  • Insilico's rentosertib, the furthest-along AI-designed molecule, only reached a Phase IIa efficacy signal.
  • AI-derived molecules pass Phase I at 80-90% but drop to about 40% in Phase II, matching historical norms.
  • The average cost to turn any drug candidate into an approved drug is still about $2.6 billion.

💬 Smart takes

  • Elias Tharakan, cited in Clinical Trial Vanguard: the industry has committed more than $7 billion to Insilico deals alone since January, with zero approvals to show for it.
  • Clinical Trial Vanguard: "the industry is funding the wrong race."
  • Skeptic on the other side: a Phase IIa efficacy signal is real progress. Traditional drug discovery took decades to reach this point too.

🧭 Where this goes

  1. Likelyat least one Insilico or Isomorphic Labs candidate reaches a Phase III readout within 18 months.
  2. Likelypharma keeps signing AI deals regardless of trial outcomes, because the deals are cheap next to $2.6B per drug.
  3. Possibleone high-profile AI drug candidate fails Phase III publicly, denting the whole category's credibility.
  4. Wild Cardthe first AI-discovered drug wins approval before the end of 2027, resetting the narrative overnight.

🥄 The Spoon Take

AI drug discovery didn't fail, it hit the same wall every drug hits: slow, unforgiving human trials. The $7 billion bets AI shrinks that wall eventually. The zero approvals say it hasn't yet. Both are true, and that's the real story, not the hype or the debunking.

🤔 Pushback

Rentosertib's Phase IIa signal is a genuine result, and dismissing all AI drug discovery as hype ignores that one real data point.

·Clinical Trial Vanguard - $7B, zero approvals·IntuitionLabs - Lilly-Insilico deal
Friday Jul 10
HEALTHOther

FDA Accepts Its First AI Liver Tool

10JUL
ACCEPTEDLIVER MODELFDA

A biotech startup just cleared a rare regulatory hurdle. Its AI-built liver model is now inside the FDA's own toolkit. Most AI drug efforts are still chasing their first approval.

Boston startup Absentia Labs is the name behind it. Its Digital Liver Model is the first AI drug-safety tool the FDA has ever formally accepted.

The tool flags drug-induced liver injury, a leading cause of failed trials. It blends liver biology with AI trained on real drug-response data. CEO Farhan Khodaee co-founded the company in Boston back in 2024.

The industry has burned $60 billion on AI drug discovery with zero approvals so far. A slow regulatory win is still a different kind of progress than another launch.

full brief & sources

⚡ Why this matters

  • It's a concrete regulatory milestone in a field known for hype without approvals.
  • Predicting liver toxicity earlier could cut years off failed drug development timelines.

🔍 What happened

  • Absentia Labs' Digital Liver Model was accepted into the FDA's ISTAND qualification pathway on July 7.
  • It's the first AI-driven Drug Development Tool accepted into that program.
  • The model predicts drug-induced liver injury (DILI), a leading cause of drug-development failures.
  • It combines mechanistic liver biology with AI trained on drug-response data.
  • Absentia Labs, founded in Boston in 2024, is led by CEO and co-founder Farhan Khodaee.
  • The milestone lands the same week industry reporting pegs $60 billion in AI drug-discovery investment against zero FDA-approved AI-originated drugs.

💬 Smart takes

  • Absentia Labs: positions the model as letting developers "assess liver injury risk earlier" in the pipeline.
  • Industry data (Clinical Trial Vanguard): roughly 175 AI-originated drug programs have entered human trials; none has an FDA approval yet.
  • Skeptic: qualification as a development tool is not a drug approval. It speeds up testing, it doesn't guarantee any drug works.

🧭 Where this goes

  1. Likelymore AI biology startups pursue the same FDA qualification-tool route instead of racing to originate drugs directly.
  2. Possiblethis becomes the template other 'AI model of biology' startups point to as proof of regulatory traction.
  3. Possiblepharma partners license the Digital Liver Model to de-risk their own pipelines rather than build in-house.
  4. Wild Cardan AI-flagged liver-injury signal kills a late-stage drug candidate publicly within the next year, becoming the tool's real proof point.

🥄 The Spoon Take

$60 billion into AI drug discovery and zero approvals is the headline everyone quotes. This is the quieter, more useful story: an AI tool got accepted into the FDA's own process, not just a lab's hype deck. Regulatory plumbing is slow and boring, and it's also how this field actually starts paying off.

🤔 Pushback

Qualification is a process step, not a result; the model still has to prove it predicts injuries the industry's existing tools miss.

·BioSpace - Absentia FDA milestone·Absentia Labs - Company site·Globe and Mail - Absentia FDA wire
Thursday Jul 9
HEALTHOther

Pharma Spends $7B, Approves Zero AI Drugs

7JUL
$7B SPENT0 APPROVED

Finding candidates got cheap and fast. Getting one approved still costs $2.6 billion and years of trials. That gap between the two is where the real money question sits now.

Drugmakers have signed over $7 billion in deals with Insilico Medicine alone since January. Not one of those partnerships has produced an approved medicine yet.

Insilico's lead compound, rentosertib, cleared Phase 2a with positive results, but Phase 2b, Phase 3, and manufacturing review are still ahead. Roughly one in eight candidates ever reaches a patient, and the full path runs about $2.6B.

Software solved the search problem. It hasn't touched the slow part: messy trial data, long timelines, and regulators who still want a human sign-off in the file.

full brief & sources

⚡ Why this matters

  • Reframes AI drug discovery hype: the hard, expensive part was never finding the molecule.
  • $2.6 billion and a 12% success rate are the real numbers that decide if this pays off.
  • More discovery deals mean more candidates competing for the same scarce trial capacity.

🔍 What happened

  • Since January 2026, pharma has committed more than $7 billion to AI-discovery partnerships with Insilico Medicine, covering Servier, Eli Lilly, SK Biopharmaceuticals, and Takeda.
  • Zero AI-discovered drugs have won FDA approval to date.
  • Insilico's rentosertib is the furthest along, with positive Phase 2a results in a peer-reviewed study.
  • Getting one drug from candidate to approval still costs about $2.6 billion.
  • Only about 12% of molecules that enter trials ever reach patients.
  • Clinical trial data across sponsors remains siloed and not model-ready, unlike molecule-design data.

💬 Smart takes

  • Milad Alucozai: "Filters eliminate garbage. They don't create gold" - AI discovery is a better filter, not a way to invent breakthrough biology.
  • Skeptic (Dr. Guy Stephens): Claude Science and similar tools are a step toward regulatory-ready AI, but the FDA still has no pathway for AI output to substitute for investigator judgment.

🧭 Where this goes

  1. Likelymore AI-discovery capital shifts toward trial operations like enrollment and site selection once sponsors feel this bottleneck directly.
  2. Likelyrentosertib's Phase 2b and 3 results become the industry's real test case for the whole category.
  3. Possiblea major sponsor publicly redirects an AI budget from discovery to clinical-ops tooling this year.
  4. Wild Cardthe first AI-discovered drug wins approval within 18 months, resetting the narrative entirely.

🥄 The Spoon Take

The industry just spent $7 billion proving it can build a faster molecule-finding machine. It didn't spend a dollar fixing the slow, siloed, trust-starved system that decides if any of those molecules ever reach a patient. Faster discovery bolted onto a broken pipeline doesn't speed up drugs. It just creates a longer line.

🤔 Pushback

Rentosertib's early Phase 2a data could still fail in Phase 3 like most candidates do, which would undercut the "AI discovery works, execution is the problem" framing entirely.

·Clinical Trial Vanguard - $7B, zero approvals·IntuitionLabs - Lilly-Insilico $2.75B AI drug deal analysis·Artificial Intelligence News - Takeda's $600M Insilico deal
Sunday Jul 5
HEALTHOther

AI Predicts Alzheimer's Ten Years Early

5JUL
AHADOCTORAI

MIT's new model spots Alzheimer's markers in brain scans a decade before symptoms show. Trials on 5,000 patients hit 87 percent accuracy. Insurers are already asking about deployment.

Standard cognitive tests catch Alzheimer's only after real damage is done. The MIT model reads MRI patterns humans cannot see, flagging risk in patients who still feel fine.

Twelve pharma companies were testing failed drugs on late-stage patients. Early detection means testing on the RIGHT patients at the right stage. The pipeline gets another shot.

Insurance is where this lands next. Should coverage change if you get flagged early? The FDA has not decided. The ethics have not caught up. The tech is ready.

full brief & sources

⚡ Why this matters

  • First AI system with a decade-forward Alzheimer's prediction validated at cohort scale
  • Unlocks a real drug-development pipeline that had stalled — trials on presymptomatic patients
  • Redefines the diagnostic starting line for a disease that affects 55 million people globally

🔍 What happened

  • MIT CSAIL model published in Nature Medicine, June 2026
  • Trained on 5,000 MRI scans linked to 20-year outcome data
  • 87% sensitivity, 91% specificity for progression to Alzheimer's within 10 years
  • Detects micro-structural changes in the hippocampus and entorhinal cortex invisible to radiologists
  • Requires only standard T1-weighted MRI — no PET or CSF markers

💬 Smart takes

  • Dr. Reisa Sperling (Harvard Aging Brain Study): 'This is the tool the field has needed for twenty years to run prevention trials.'
  • Skeptic read: 87% sensitivity means 13% miss. And what does an early flag DO for someone who has no treatment options yet?
  • Insurance analyst read: if the flag becomes standard-of-care, expect a serious debate about pre-existing condition rules for long-term care coverage.

🧭 Where this goes

  1. LikelyNIH funds the first prevention-drug trial using the model for enrollment by end of 2026
  2. Likelypharma companies restart shelved Alzheimer's programs based on early-stage cohorts
  3. PossibleFDA issues 510(k) clearance within 18 months for clinical deployment
  4. Wild Cardan insurer starts adjusting long-term care premiums based on early flags — triggers a policy firestorm

🥄 The Spoon Take

The medical AI story we have been waiting for. Prediction without treatment sounds cruel — but early flags are exactly what the drug pipeline needs. This one moves the whole field forward.

🤔 Pushback

Detection without a proven intervention still puts patients in a strange middle zone. And 87% accuracy on brain data is not the same as 87% accuracy on prognosis at the individual level.

·MIT CSAIL preprint in Nature Medicine