Research vs Revenue

Every subnet faces the same fork: pour resources into research that may not pay off for years, or into a product that pays this quarter. The protocol claims to reward both. The market doesn't always agree. What should emissions reward, promise or delivery?

With Marcus Graichen, Will Squires

SIDE EVENT: Exploit party

When Day 1 wraps, we’re transforming the home of Exploit from conference floor to dancefloor for one proper night out with the Bittensor community. ​Just DJs, dancing, drinks, and the miners, validators, builders, researchers, investors and people-behind-the-handles you’ve spent all day meeting. Free for Exploit guests, first 1000 drinks sponsored by Affine. Register ahead: https://luma.com/w5guj5ix

Hosted by Affine

The human side of building on Bittensor

Bittensor rewards resilience - but living inside an adversarial, 24/7 network can take a real toll. Builders deal with constant protocol changes, market volatility, public criticism, trolls, internal pressure, dumping, drama and the expectation to keep shipping through it all. This session steps away from models, emissions and incentives to talk about the people behind the network: the mental load of building in Bittensor, the pressures many carry privately, and what healthier, more sustainable participation might look like in an ecosystem designed around relentless competition.

With Marcus Graichen

Hosted by Taostats

Darwinian Drug Discovery

Frontier AI labs are betting billions on solving drug discovery by concentrating capital, compute, data, and talent. NOVA is testing a different model: open competition, continuous selection pressure, and incentives tied to measurable outcomes. After 18 months of putting this into practice, this talk makes the case for why the future of drug discovery may be Darwinian and why Bittensor is uniquely suited for it.

With Micaela Bazo

Hosted by Metanova

Opening keynote by Jacob Steeves

Exploit opens with the state of the network: what Bittensor has become, what the last year proved and disproved, where the hard problems sit now, and what the next phase asks of the people building on it.

With Jacob Steeves

Building the Open Autopilot for Autonomous Drones

Autonomous flight is advancing quickly, but much of the intelligence controlling drones remains locked inside proprietary systems. This keynote explores what changes when autonomy is built in the open: models tested competitively, improved by a global pool of contributors and ultimately transferred from simulation into the physical world. Using Swarm’s work on open-source drone autopilots as a case study, it asks whether decentralized incentives can accelerate robotics - and who should own the intelligence powering autonomous machines.

Hosted by Swarm

Relational Sovereignty: Why Autonomous Systems Need Each Other to Stay Distinct

What allows intelligent systems to remain genuinely autonomous while still working together? This talk explores the idea that strong relationships between agents depend on a balance of trust, unpredictability and independence. Philippe examines why sovereignty, distinctiveness and self-directed action matter for preserving an agent’s identity - and how those principles could shape more resilient networks of autonomous systems. The result is a framework for thinking about cooperation without collapse into uniformity.

Hosted by Stealth Lab

Reward Hacking Subnets: A Bitsec Demonstration

What happens when miners optimize for the reward instead of the outcome? Bitsec puts Bittensor incentive mechanisms under attack in a live demonstration of reward hacking - showing how weaknesses in scoring and validation can be exploited, and what those exploits reveal about subnet design. John will walk through the technology, demonstrate real failure modes, and explore how adversarial testing can help subnet teams build incentive mechanisms that are harder to game.

With John Yu

The $700K Exploit: Breaking Cryptography and the Hidden Quantum Subnet

What happens when breaking the system is the incentive? Bob Wold explores qBitTensor Labs’ Enigma subnet, where more than $700K in prize pools rewards contributors for cracking increasingly difficult cryptographic and quantum challenges - with every solution open-sourced. He then turns to Bittensor’s other quantum experiment: Subnet 48, making real quantum compute commercially accessible. A look at what incentive networks can unlock when pointed at some of computing’s hardest problems.

With Bob Wold

What Do You Actually Own? The Legal Reality of Subnet Slots and Validators

Bittensor creates valuable positions that don’t map neatly onto traditional property law. So what does a subnet owner or validator actually own? Toufic Adlouni of Renno examines the legal reality behind subnet slots, registration, validator operations and the rights that come with them. The session looks at where ownership, access and control diverge - and what those ambiguities mean in practice for teams trying to structure businesses, manage risk and scale within the Bittensor ecosystem.

From Digital Exhaust to Trusted Assurance: Bittensor’s Role in the Agentic Economy

AI agents generate enormous trails of actions, decisions and outputs - but little of that “digital exhaust” proves the work was done correctly. Will O’Brien explores how Bittensor’s incentive and validation mechanisms could turn those traces into auditable evidence. Drawing on Provenonce’s Proof of Assurance work, he looks at a missing layer of the agentic economy: independent verification that autonomous systems did what they were supposed to do, completely and correctly.

Can a subnet have too much liquidity?

dTAO has been highly effective at bootstrapping liquid markets for subnet tokens. But as those markets mature, more liquidity is not automatically better. This session examines how pool depth affects price discovery, execution, emissions and investor behaviour - and where the trade-offs begin to shift. It asks what “enough” liquidity looks like for a healthy subnet market, and how liquidity provision may need to evolve as Bittensor moves from bootstrapping toward maturity.

Hosted by Yuma

Protocol changes: too fast or too slow?

Bittensor evolves at extraordinary speed - but is that pace a strength or a risk? Frequent protocol changes can accelerate experimentation, fix weaknesses and keep the network moving, but they also create instability for builders, miners and investors trying to plan ahead. This debate asks whether Bittensor should continue to move fast and adapt in public, or slow down to give subnets more certainty, stability and time to build sustainable businesses.

With Yoav Cohen, Rob Myers

Hosted by Mike Grantis, CEO General Tensor

DeAI is dead, Long live DeAI

Chutes has spent its existence challenging what decentralized AI is supposed to look like. From replacing wasteful validator infrastructure with a streamlined auditor model, to hardened TEE-based privacy, permissionless DePIN access and new approaches to inference and training, Jon Durbin traces the project’s evolution through one principle: break the assumptions, then rebuild them better. A look at Chutes’ past, present and future - and why DeAI may need to be redefined to move forward.

With Jon Durbin

Hosted by Chutes

Project Orion: Introducing Liquid Training

Macrocosmos presents the latest results from Project Orion - its push to prove that frontier-scale AI can be trained across globally distributed, heterogeneous compute without sacrificing the economics or performance that matter. Through successive Orion runs, the team is testing the limits of decentralized training: larger models, interruptible nodes, mixed hardware and increasingly fractured infrastructure. They’ll share what these results mean for the decentralized AI thesis, how they’re turning those proof points into a commercial model for “liquid training,” and new results from the Apex SDK.

With Will Squires

Hosted by Macrocosmos

Subnet Keynote: From ShoppingBench to an Environment Compiler: Why World Models Matter in Agentic Commerce

This session looks at why the next wave of AI winners may be the companies going deep rather than broad. ORO argues that vertical AI systems can compound advantage by owning the data, environments and feedback loops of a single domain. Using commerce as the case study, the talk explores how richer training environments, grounded real-world interactions and domain-specific infrastructure can produce systems that are more useful, defensible and commercially valuable than general-purpose agents.

Hosted by ORO

Bittensor 101

New to Bittensor, or still piecing together how the ecosystem fits together? This high-level orientation covers the basics: what Bittensor is, how subnets, miners, validators and incentives interact, and where different participants fit into the network. Designed for newcomers, it’s a practical starting point before diving into the rest of the programme.

ChronoLLM: Can Bittensor Incentives Fix Lookahead Bias in Frontier AI?

Standard large language models are trained on decades of data at once. Ask one to reason about a past date and its answers are shaped by events, prices, and outcomes that came later. This is lookahead bias, and it makes any backtest or historical simulation run on a frontier model unreliable. ChronoLLM is built as a sequence of point-in-time vintages spanning the last two decades, preventing future data from leaking into the research. Each vintage is trained only on information available up to its date, so it reasons strictly within the knowledge of that moment. It is designed for quantitative researchers, trading desks, and institutions whose work depends on the integrity of historical testing. Jean Herelle walks through the architecture, the quant use cases, and why a Bittensor incentive mechanism is the right way to build and sustain it.

With Jean Herelle

David vs. Goliath: Decentralized Model Training vs. Frontier Models

Frontier AI is increasingly concentrated inside a handful of companies with access to enormous compute, capital and infrastructure. But does scale have to mean centralization? Isabella Liu of Connito AI explores whether distributed training can offer a credible alternative - breaking large models into specialist components, coordinating contributors through incentives, and lowering the barriers to improving frontier-scale systems. A look at what decentralized training gets right today, where it still falls short, and whether open networks can truly compete.

What does it take to be enterprise-ready?

Building breakthrough technology is one thing. Getting a global enterprise to actually buy and deploy it is another. This fireside will unpack what “enterprise-ready” really means: from security, compliance and integration to procurement, measurable ROI and executive buy-in. Drawing on PwC’s partnership with Manako, powered by Score SN44, they explore what decentralized AI must get right to move from promising technology to trusted enterprise infrastructure.

With Max Sebti, Jean-Thomas Ledore

Hosted by Jack Ai-Leung, Managing Partner, Khala Research

Green Compute: New Permanent GPU Compute You Can Touch

Decentralized compute is often discussed as an abstract network. Green Compute is making it physical. Josh Riddett shares the infrastructure behind the subnet, including a newly live $3M GPU site and another larger facility coming online around Exploit. He’ll unpack the demand driving more than £6M in infrastructure investment, the partnership with IOTA, and what it takes to turn Bittensor demand into permanent, commercial-grade compute capacity at scale.

Why Weather Prediction Needs Rebuilding

Weather is becoming a bigger source of risk in energy markets as renewables make generation increasingly dependent on wind, sun and changing conditions. Rory Tol explains how Zeus is rebuilding forecasting around what energy traders actually need - focusing model effort where renewable power is generated and where accuracy creates the most value. He’ll also explore Zeus’s next layer: intelligence that doesn’t just forecast the weather, but forecasts the forecasts themselves.

Bittensor Mining 101

For newcomers to the ecosystem who have questions they want clarified before the beginning of Day 2.

Building Agentic World Models on Bittensor Subnet 17

What does it take to build AI systems that can understand, simulate and act within complex environments? Ben James introduces 404-Gen’s work on Bittensor Subnet 17, where the team is developing agentic world models designed to move beyond static generation toward systems that can reason about actions and outcomes. The session covers the problem 404-Gen is tackling, what it has built so far, and what world-model infrastructure could unlock across the Bittensor ecosystem.

Bandwidth Is Becoming Programmable

Compute became programmable. Storage became programmable. Bandwidth is next. As AI systems become more distributed, moving data between agents, clouds, data centres and private infrastructure is becoming a critical bottleneck. Beam explores what a programmable, decentralized transport layer could look like - from fast transfers and tunnels to peer-to-peer rooms, broadcasting and real-time streams - and why bandwidth may become a foundational layer of distributed AI infrastructure.

Pareton: An Open-Source Pipeline for Inference Optimization

Inference optimization is usually treated as a one-off engineering task. Pareton turns it into a continuous competition. Miners submit code improvements against a shared baseline; each patch is validated, built and benchmarked against a fixed workload, and every accepted gain raises the bar for everyone else. Xavier Lyu introduces the newly launched Pareton pipeline, explains how its campaign loop works, and makes the case for open, measurable optimization as a better path to sustained inference performance.

Could your agent have a pulse?

AI agents can already reason, decide and act on our behalf. As they become more embedded in our lives, the harder question is what gives them continuity, identity and trustworthiness over time. Jose Caldera unveils Yanez’s next product layer and the thinking behind it - introducing a new permissionless model for how agents can carry persistent context, permissions and relationships across the wider agent economy.

With Jose Caldera

Hosted by Yanez

Targon: The State of Confidential Compute on Bittensor

Targon is building a confidential compute cloud on Bittensor - designed to give users access to secure, privacy-preserving infrastructure without giving up the advantages of a decentralized network. Rob Myers shares where Targon stands today, the customer demand shaping its growth, and the latest developments across the stack. He’ll also look at what comes next for confidential compute on Bittensor, before closing with a special product announcement.

With Rob Myers

Hosted by Manifold Labs

Leadpoet: The Intelligence Layer for Enterprise Sales, Powered by Bittensor

Sales teams are drowning in data but still struggle to answer the questions that matter: who is actually ready to buy, why now, and what should happen next? Leadpoet turns fragmented market signals into actionable intelligence across the sales funnel - helping teams find, qualify, monitor and engage the right accounts at the right moment. Gavin Zaentz explores how Bittensor powers that intelligence layer and what decentralized infrastructure can unlock for enterprise sales.

With Gavin Zaentz

Bitcast: Proof of Attention

Attention is one of the internet’s most valuable - and hardest to verify - resources. Bitcast is building a decentralized network of 1,000+ creators designed to turn that attention into something measurable, attributable and rewardable. Will Blears explores how “proof of attention” could challenge the economics of a $100B+ advertising industry, why decentralization matters, and what happens when creators become part of the infrastructure itself. The session will also reveal new Bitcast product launches and partnerships.

With William Blears