From openly blocking models for relevant pharma problems, to silently reducing Fable capabilities for frontier AI research, to now fully blocking the use of Fable, the signal is clear.
There has never been a more relevant and exciting time to be building in the open, right at the critical intersection of AI and drug development. This is where incentivised open science will lead.
#Bittensor
Decentralized AI isn't just philosophically better, it's functionally necessary for science to move at the speed required.
Anthropic already made Claude structurally useless for serious biotech work before any government pressure.
When they launched Claude Opus 4, they activated ASL-3 — their internal Responsible Scaling Policy's highest deployed tier. The trigger? Internal testing showed the model could assist someone with a basic STEM background in synthesizing dangerous pathogens. Their chief scientist called it explicitly: COVID-like agents, pandemic-level risk.
Their response was to apply broad biological research restrictions across the board.
The problem is that the same guardrails that block novice bioterrorists also block legitimate computational drug discovery, protein engineering, nanobody design, and molecular pathway analysis. You can't surgically remove "dangerous biology" from "useful biology" at the model level — the underlying science is the same. So they blunted the whole thing.
This is what centralized AI control looks like in practice. One company's risk tolerance — shaped by liability, regulatory pressure, and investor optics — becomes the ceiling for what an entire scientific field can access.
Open source isn't just an ideological position — it's infrastructure for the next era of drug discovery.
Amazing to talk to so many cracked devs about what we are doing at @metanova_labs during the bittensor ideathon happening on the MuShanghai @themu_xyz just after joining the amazing biotech and longevity week organized by our partners from @Yalotein
🔥
Exciting to see the live-competition virtual screening systems we built at @metanova_labs connect with @onepot_ai’s automated robotic chemical synthesis platform - bringing the design–make-test-learn loop closer to real time
Excited to announce our partnership with https://t.co/uIdhcF7Hpb
@onepot_ai built the AI-driven robotic synthesis lab we've always wanted on the other side of our screening pipeline, enabling shipments of novel small molecules from their fast-growing CORE collection in as little as 5-7 business days rather than months.
Founders @daniil_boiko and @andrei_tyrin raised $13M from @fiftyyearsvc, @khoslaventures, @Speedinvest, @norrsken_vc, and @script_cap, with support from Agata and @woj_zaremba (co-founder of @OpenAI), @JeffDean (Chief Scientist at @Google), and @NAVEC to end the synthesis bottleneck in drug discovery.
This partnership will turbocharge @metanova_labs R&D. Hits identified with the help of NOVA’s screenings and algorithms (#Bittensor #SN68) can now move to validated physical compounds faster than ever.
This is the foundation for something bigger: a path toward a fully agentic discovery loop.
It also opens up something new.
Partners can now access decentralized virtual screening + robotic synthesis as a coordinated service.
The future of drug discovery is autonomous.
We're building toward it, one integration at a time.
#DrugDiscovery #Biotech #AI #DeSci #Robotics
NOVA Nanobodies competition launched 2 weeks ago.
3,000 submissions. Data got better fast!
How submissions improved over time:
- Binding confidence: up ~3x from first submission to last
- Structural fit: uncertainty at the interface dropped by nearly half
- Hydrogen bonds: grew steadily across all 3,000 designs with no sign of plateauing
- Buried surface contact: increased modestly but steadily, with later designs gripping the target across a larger area
- Salt bridges: flat early, then accelerated sharply after submission ~2,000
The target: PD-L1, the protein cancer uses to hide from your immune system. Blocking it is the mechanism behind some of the most successful cancer drugs ever approved like Keytruda, Opdivo, Tecentriq.
Nanobodies are a next-generation format: smaller, cheaper to manufacture, and better at penetrating tumors than conventional antibodies. A well-designed anti-PD-L1 nanobody could become a more accessible cancer immunotherapy, or a building block for next-generation bispecific drugs that hit 2 targets at once.
This happened permissionlessly.
No central lab.
Just an open incentive mechanism rewarding the best work.
#Bittensor #SN68 #DeSci #DrugDiscovery #Cancer #Nanobodies
ArboNOVA: Patent–Molecule matching loop
We’ve been experimenting with an agent that maps molecules → prior art using only open data + tools
Benchmark:
~1500 molecules across ADHD-related patents (since 2012)
In ~12 hours:
18 iterations of the loop → Best hit rate: 85.4%
How this is usually done:
Pharma intelligence teams + expensive proprietary databases + manual workflows + even conference attendance
Early, but promising.
Moving one step closer toward automating drug discovery and identifying which molecules are most strategic to advance in the wet lab.
Based on @const_reborn (https://t.co/RgNwM6udgv) and @karpathy autoresearch framework
#Bittensor #SN68 #ralphloop #agents #DrugDiscovery #Desci #DeAI
AI4Science makes me think about Popper vs. Kuhn. I'm really excited for progress in the Popperian sense, and have not yet seen progress in the Kuhnian sense.
Popper saw science as progress through rapid cycles of hypothesis → experiment → falsification. AI is an incredible engine for that: faster predictions, cheaper tests, tighter feedback loops. I would say the vast majority of work in science and engineering falls into this category; thus, if we get AI for science right, this area should lead to big changes in how science is conducted broadly.
Kuhn argued that the biggest breakthroughs are paradigm shifts that change the entire set of questions and concepts (Newton, Darwin, transformers, quantum mechanics). Importantly, this is how real progress is made and accounts for a small proportion of the effort in science. I don't think we've seen this yet by AI. I'm not sure how far away we are from seeing beautiful experiments that change the way we think, proposed by AI.
NOVA Compound's new scoring function builds upon @MIT and @RecursionPharma Boltz-2's predictions to improve BOTH Hit and Hit-to-Lead identification!
Better predictions means less $$$ wasted in experiments downstream.
Read more about how we evaluated this new approach in the article below.
Metanova Labs x DiaGen AI Inc
announcing new JV to develop a multi-objective Hit Picker for automating drug discovery
today, Hit-picking is slow and manual with medicinal chemists reviewing ranked lists and relying on years of experience and intuition.
together we’re building a tool that can automate this step by evaluating small molecules for novelty, affinity, and target specificity across multi-million and billion-scale libraries.
the Hit Picker will be used to select and curate the best NOVA submissions into token-gated compound libraries for external biotech and pharmaceutical clients.
this collaboration scales the way we can identify and prioritize top candidates for validation, transforming early discovery throughput and value.
🔗 read the full press release: https://t.co/AFk3IOMxr0
#bittensor#SN68 is evolving
introducing NOVA Blueprint: an open algorithmic challenge to innovate faster, smarter molecular search methods for ultra-large datasets that can plug into any model for any property.
this will be released as a new incentive layer inside the subnet thanks to recent updates to the chain. NOVA Blueprint turns the discovery process itself into an ever improving process with the power of tokenized incentives.
test it, break it, give us your feedback.
now live for testing:
https://t.co/OYfYt5UPBU
$NOVA is now featured in 4 of @TrustedStake automated strategies for maximizing $TAO returns:
-TrustedStake Blue Chip Subnet Index
-Doug's (@cisterciansis) picks
-TSDCA Blue Chips
-Bittensor Universe
🧬💊AI for biology + chemistry still lags behind today’s frontier language models
but there is so much reason for hope, if you are willing to look for it
one day this winter will end, and the world will realize the golden goose they have been ignoring ❄️🦢🪙
The largest updates until now on our mission for alignment between all parts involved in decentralizing virtual screening on the #bittensor ecosystem
#sn68#tao