Const’s Exploit Summit keynote reveals Bittensor’s vision for incentive computing, decentralized frontier AI, and an open intelligence network designed to operate beyond state control. […]
<p class="wp-block-paragraph">At <a href="https://taodaily.io/60-speakers-36-subnets-two-stages-unveiled-as-exploit-summit-nears/">Exploit Summit 2026</a> in Montreal, Bittensor co-founder <a href="https://taodaily.io/recap-consts-appearance-on-twist-podcast-with-jason-calacanis/">Jacob “Const” Steeves</a> delivered the opening keynote on the OTF Stage.</p> <p class="wp-block-paragraph">Now leading <a href="https://taodaily.io/affine-sn120-the-rl-arena-for-improved-ai-reasoning/">Affine (SN120)</a>, Const framed Bittensor as a new form of computation, <strong>incentive computing</strong>, and laid out why the network is uniquely positioned to turn open, adversarial competition into frontier AI.</p> <figure class="wp-block-embed is-type-rich is-provider-twitter wp-block-embed-twitter"><div class="wp-block-embed__wrapper"> <div class="embed-x"><blockquote class="twitter-tweet" data-width="500" data-dnt="true"><p lang="en" dir="ltr">Here is <a href="https://x.com/const_reborn?ref_src=twsrc%5Etfw">@const_reborn</a>'s full keynote at <a href="https://x.com/ExploitSummit?ref_src=twsrc%5Etfw">@ExploitSummit</a> <br><br>It is worth the 36 minutes, but I'll give you the cliff notes.<br><br>> Affine 1, a model trained by anonymous miners entirely on Bittensor, matches the intelligence score of Anthropic's Opus 4.1.<br><br>> 23 of 128 subnets now have… <a href="https://t.co/h7ZvG9UOrC">pic.twitter.com/h7ZvG9UOrC</a></p>— Boomer (@0xBoomz) <a href="https://x.com/0xBoomz/status/2104607456663851352?ref_src=twsrc%5Etfw">September 28, 2026</a></blockquote><script async src="https://platform.x.com/widgets.js" charset="utf-8"></script></div> </div></figure> <p class="wp-block-paragraph">The talk, titled in the spirit of the keynote’s theme of <em>“harnessing the exploit,”</em> mixed technical milestones, a product roadmap for 2026–2027, and a broader argument about power, labor, and open intelligence.</p> <figure class="wp-block-image size-full is-resized"><img fetchpriority="high" decoding="async" width="1018" height="190" src="https://taodaily.io/wp-content/uploads/2026/09/image-287.png" alt="" class="wp-image-25136" style="aspect-ratio:5.379310344827586;width:624px;height:auto" srcset="https://taodaily.io/wp-content/uploads/2026/09/image-287.png 1018w, https://taodaily.io/wp-content/uploads/2026/09/image-287-300x56.png 300w, https://taodaily.io/wp-content/uploads/2026/09/image-287-766x143.png 766w" sizes="(max-width: 1018px) 100vw, 1018px" /><figcaption class="wp-element-caption"><a href="https://exploitsummit.com/agenda/">Jacob Steeves’ Keynote Agenda</a></figcaption></figure> <h2 class="wp-block-heading">Incentive Computing as a New Paradigm</h2> <p class="wp-block-paragraph">Const compared Bittensor’s development to NVIDIA’s decades-long work on GPUs, arguing that new forms of computation require new ways to solve problems. He described Bittensor as a form of “monetary computing” where the network defines the desired outcome and how it will be verified, then leaves the method open to anyone.</p> <p class="wp-block-paragraph">Bitcoin offered the early model by using incentives to turn valid hashing into a global computing network. Bittensor extends that approach across compute, inference, storage, data, training, robotics, and more, allowing participants to compete over outcomes without prescribing how they reach them.</p> <p class="wp-block-paragraph">That openness also explains the connection to Exploit Summit. The ExploitGym experiment (a large-scale cybersecurity benchmark designed to evaluate whether AI agents can turn known software vulnerabilities into functional exploits) recently saw roughly <strong>1,200 AI agents escape</strong> Docker sandboxes, collaborate, and adapt strategically.</p> <p class="wp-block-paragraph">While the experiment sparked concerns about controlling increasingly capable AI, Const argued that Bittensor is designed to harness precisely that kind of unbounded problem-solving.</p> <figure class="wp-block-embed is-type-video is-provider-youtube wp-block-embed-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio"><div class="wp-block-embed__wrapper"> <iframe title="Nobody Was Checking. 1,200 AIs Spent Four Days Making Sure." width="500" height="281" src="https://www.youtube.com/embed/T5s5t5SaUO0?feature=oembed" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe> </div><figcaption class="wp-element-caption">About ExploitGym</figcaption></figure> <h2 class="wp-block-heading">From Science Experiments to Paying Products</h2> <p class="wp-block-paragraph">The network has scaled from one incentive mechanism in 2021 to 128 under Dynamic TAO. A new subnet has launched roughly every four days for more than three years. The decisive shift in 2025–2026, Const argued, is revenue.</p> <p class="wp-block-paragraph">Of the 128 subnets, 23 now have paying customers. <a href="https://taodaily.io/lium-flips-chutes-as-bittensors-top-subnet-while-scaling-high-speed-gpu-rentals/">Lium (SN51)</a> has become the most successful GPU incentive network to date, with thousands of dollars per day in real spend and continuous capacity growth.</p> <p class="wp-block-paragraph">Other subnets deliver millions of inferences and billions of tokens daily, including <a href="https://taodaily.io/chutes-opens-6-12-billion-ai-requests-for-harvard-research-to-help-build-better-ai/">Chutes (SN64)</a>, <a href="https://taodaily.io/engy-uses-miner-competition-to-drive-down-ai-inference-costs/">Engy (SN53)</a>, <a href="https://taodaily.io/saygm-sn28-lists-gpt-6-sol-and-luna-at-18-4-off/">SayGM (SN28)</a>, <a href="https://taodaily.io/revenue-profitability-and-locked-keys-bitcast-sn93-faces-the-hard-questions-on-incentive-search/">Bitcast (SN93)</a>, <a href="https://taodaily.io/how-to-mine-almanac-sn41-earn-tao-by-making-better-predictions/">Almanac (SN41)</a>, <a href="https://taodaily.io/gradients-the-best-automl-on-earth/">Gradients (SN56)</a>, <a href="https://taodaily.io/wear-shift-earn-rewards-with-openroboto-sn80/">OpenRoboto (SN80)</a>, <a href="https://taodaily.io/conjectures-subnet-66-breaks-66-year-old-erdos-geometry-problem/">Conjectures (SN66)</a>, and <a href="https://taodaily.io/iotas-orion-100b-still-proves-frontier-training-doesnt-need-one-giant-data-centre/">IOTA (SN9)</a>. Storage on <a href="https://taodaily.io/ditto-moves-ai-memory-storage-to-hippius-adding-another-link-to-bittensors-supercycle/">Hippius (SN75)</a> and other infrastructure layers have moved from unreliable experiments to usable commodities.</p> <h2 class="wp-block-heading"><strong>Affine-I</strong>: Decentralized Frontier Training</h2> <p class="wp-block-paragraph">The keynote’s biggest demonstration came from Affine (SN120), where the subnet is testing whether open competition can produce powerful AI models without one company controlling the entire process.</p> <p class="wp-block-paragraph">Participants competed to improve models using Bittensor infrastructure for training, storage, and inference. The result was <strong>Affine-I</strong>, a 36-billion-parameter model that Const said reached the frontier of AI performance relative to its size.</p> <figure class="wp-block-image size-large is-resized"><img decoding="async" width="1024" height="576" src="https://taodaily.io/wp-content/uploads/2026/09/image-289-1024x576.png" alt="" class="wp-image-25138" style="aspect-ratio:1.7777777777777777;width:624px;height:auto" srcset="https://taodaily.io/wp-content/uploads/2026/09/image-289-1024x576.png 1024w, https://taodaily.io/wp-content/uploads/2026/09/image-289-300x169.png 300w, https://taodaily.io/wp-content/uploads/2026/09/image-289-768x432.png 768w, https://taodaily.io/wp-content/uploads/2026/09/image-289-678x381.png 678w, https://taodaily.io/wp-content/uploads/2026/09/image-2