Media room

Without verification, society is forced to lean on institutional trust, corporate assurances, and narrative reconstructions drawn from mutable logs. With verification, it can lean on evidence: cryptographic, reproducible, and specific to each decision.

Inference Labs builds a proof-of-inference layer that plugs into existing vision pipelines – claims triage systems, rental damage scanners, platform trust stacks – without forcing teams to retrain or redeploy their models.

A better explanation of what we’re building at Inference Labs and why it matters, it's important to understand quantum computing threats and how they relate to zero-knowledge proofs (ZKPs) and zero-knowledge cryptography.

Meet the rebranded Subnet: the world’s fastest zkML proving cluster. Verifiable oracles for any computation via zero-knowledge circuits on Subnet 2.

Computer vision is moving from analytics to action. Through this partnership, Score’s real-time vision models can now be verified at the inference level — ensuring that detection outputs used in markets, automation, and decision systems are provably correct.

Inference Labs has teamed up with OpenLedger to develop infrastructure that ensures AI systems are transparent, auditable, and trustworthy.

Most LLM benchmarks assume the world is static. Reality isn’t. In TruthTensor, we evaluate models in live prediction markets under drift and show that models can follow instructions locally while violating them over time. Same prompts. Same data. Very different reasoning behaviour.

2025 marked the moment verifiable AI became real infrastructure, with milestones like Subnet-2 surpassing 300M+ verifiable inferences, the launch of DSperse, JSTprove, and TruthTensor. This is what it looks like when research becomes reality—proving AI can be accountable.

Inference Labs is proud to collaborate with VentureVerse to bring verifiable AI to real-world prediction markets, establishing a new foundation of trust for autonomous market systems.

In this interview, Inference Lab's co-founder, Colin Gagich, dives into verifiable inference, cryptographic guarantees, and why trust is the missing layer for physical AI.

As AI becomes more embedded in mission-critical infrastructure, unverifiable autonomy is no longer sustainable. Explore how auditable autonomy is creating trust in complex AI systems.

Learn about Hitchbot, a robot whose experimental journey went on to shape our co-founder Colin Gagich's perception of trust and it's significance when dealing with AI outputs.

TruthTensor achieved 20,000+ agents trading, 20,000+ users building agents, and 300,000+ agent decisions in its first 6 days since launch.

Explore accountable autonomy at airports, applying verifiable AI and Proof of Inference to ensure transparent, cryptographically verified decisions in high-stakes aviation environments.

TruthTensor reached 50,000+ agents trading, 40,000+ users building agents, and 400,000+ agent decisions in its first 11 days since launch by December 15, 2025.

In this exclusive interview, Ron Chan, Co-Founder & CEO of Inference Labs, reveals the origin story behind one of the most important problems in modern AI: verifiable inference.

Inference Labs has officially launched DSperse 2.0, slicing models into DSIL files for independent compilation, execution, and verifiable proofs to scale better verification pipelines with reliability from commits.

Introducing TruthTensor – a platform from Inference Labs that enables users to easily launch their own AI agents to analyze and take positions on Polymarket.

TruthTensor is live, enabling users to deploy AI agents in seconds to trade on Polymarket!

Subscription-based humanoid robots like NEO promise powerful in‑home assistance but also shift ownership and intelligence into centralized cloud services, creating serious challenges around user control, privacy, and trust.

Verifiable AI is rapidly becoming a practical requirement across regulated, high‑stakes industries.

Inference Labs' Subnet-2 (SN2) surpassed 300 million zk proofs on November 13, 2025, advancing verifiable and accountable AI autonomy through distributed verification.

Hitchbot's Guide to the Galaxy A community built around the Inference Network, its mission, its technology, and the people shaping it.

Discover DSperse, the innovative solution revolutionizing Zero-Knowledge Machine Learning (zkML) proofs through advanced model slicing.

Zero-knowledge (ZK) and artificial intelligence (AI) have increasingly complemented each other to meet growing demands for privacy, transparency, and trust in digital systems.

Inference Labs is excited to announce the public release of JSTprove, our framework for Zero-Knowledge Machine Learning (zkML).

Inference Labs is excited to announce the public release of JSTprove, our framework for Zero-Knowledge Machine Learning (zkML).

What if every AI model could prove its own output? Meet DSperse, a new open-source toolkit from Inference Labs.

AI content watermarks intended to authenticate AI-generated text, images, video, and more, can be easily removed or forged by tools and adversarial attacks, undermining their reliability.

While large language models (LLMs) generating false AI model outputs (i.e., hallucinations) are a known problem in AI systems, the deeper and more serious challenge is the systemic lack of verifiability.

Today we’re introducing something completely new: DSperse. It’s our framework for unbreakable AI — bringing strategic verification to machine learning. learn more from our whitepaper.

Meet DSperse, a modular framework that addresses this by targeted verification: rather than circuitizing an entire model, DSperse lets you circuitize and prove high-value subcomputations.

No matter how cutting-edge your AI, without provable, cryptographic trust, growth stalls dramatically and opportunities vanish.

Colin Gagich shared his founder journey — from hitchBOT to building verifiable AI at Inference Labs. The energy in the room was amazing, and the support from Toronto’s tech community was truly inspiring.

AI benchmark leaderboards are cluttered with cherry-picked and sometimes fabricated results, eroding trust in reported performance. Understanding why this matters reveals how distorted metrics can be misleading.

A few weeks ago, our co-founder Colin spoke at Google’s ZK & AI Summit. He shared why full-model zkML is hitting practical limits — and how we’re working on faster, lighter ways to solve this issue. Here’s a clip from his talk — with a glimpse of what’s coming next.

AI faces a growing credibility crisis — silent model swaps and opaque management are eroding trust when it matters most. Subnet 2 introduces real-time cryptographic verification, setting a new benchmark for dependable, accountable open-source AI.

Reddit’s UK age verification relies on unverifiable trust, exposing sensitive data to third parties. True confidentiality requires verifiable Proof of Inference and zero-knowledge systems.

Our focus in August remains the same: scaling zkML proof volumes, refining distributed proving infrastructure, making verification practical for builders.

Zero-knowledge proofs (ZKPs) improve the verifiability of media authenticity, ensuring content remains unaltered while preserving privacy.

Your AI tool suddenly becomes slower and less accurate — no notice, no explanation, just a creeping sense that it’s no longer what you paid for.

The recent strategic investment round featured follow-on participation from DACM and Delphi Ventures alongside new investments from Arche Capital and Lvna Capital. Read more details in Decrypt.

ZKPs exponentially improve the long term viability of AI, with the global zk-Proof market projected to reach more than $10 billion by 2030.

If AI is going to shape our systems, it must also prove its reasoning. That’s what Proof of Inference provides. Read the full article by our Co-Founder Ron Chan in The Defiant.

At Inference Labs, we're building networks where Proof of Inference, integrity, and incentives are foundational — and our reasoning is as open as our circuits. We conducted a Q&A with Spencer, Colin, and Eric.

AI agents are transforming digital systems by automating market analysis, optimization, and security, but they face computational limitations.

Get insights on how Inference Labs and AR.IO Are Advancing Permanent Storage Systems for AI. The value of AI inference hinges on the integrity of the data it processes.

In this emerging era, privacy-centric protocols and distributed applications are setting the stage for breakthroughs in identity authentication, all without revealing sensitive details. Dive deeper on our Medium.

We turned a Doom-playing AI agent into a verifiable system using zk-proofs—proving each move is honest, strategic, and tamper-proof in near real-time. Watch the demo and read the insights on our Medium.

Watch Colin Gagich, who is building the “SSL for AI models,” highlights how quickly top AI talent is joining the ecosystem.

Ron Chan joins Sahara's AI panel at TOKEN2049 Singapore to discuss verifiable AI and future-ready infrastructure. Watch the full session now on YouTube.

Inference Labs has closed $2.3 million in pre-seed funding as it looks to make AI inference more accessible. The round was co-led by three investors: Digital Asset Capital Management, Mechanism Capital, and Delphi Ventures.

Watch our latest clip with cofounder Ron to learn how we're scaling AI verification.