Conjecture
Conjecture is a London-based AI safety startup founded in 2022 by Connor Leahy, Sid Black, and others from the EleutherAI open-source AI research community.
Profile
Conjecture builds AI systems using a modular architecture called Cognitive Emulation that mimics human expert reasoning to improve safety and reliability.
Conjecture is a London-based AI safety startup founded in 2022 by Connor Leahy, Sid Black, and others from the EleutherAI open-source AI research community. The company develops a proprietary AI architecture called Cognitive Emulation, which aims to build AI systems that follow human-like reasoning processes to improve reliability, interpretability, and control. Conjecture has raised approximately $150 million in funding from investors including Index Ventures, Factorial Funds, and angel investors such as Jaan Tallinn, according to Crunchbase and press reports from 2023.
The company has not disclosed revenue or customer numbers as of mid-2026, and its website lists only one open job posting for a product role in London. Conjecture's core product, Cognitive Emulation, is designed to decompose complex tasks into reusable, human-expert-guided pipelines, addressing common AI failures like hallucination, incoherence, and lack of interpretability. The company positions itself as an alternative to scaling large language models, arguing that larger models amplify safety risks.
Conjecture has published research on mechanistic interpretability and AI alignment, and maintains an information hazard policy that restricts publication of certain safety-critical findings. The company has not announced any named enterprise clients or publicly disclosed partnerships. In 2025, Conjecture was notably absent from TechCrunch's list of 55 US AI startups that raised $100M or more, reflecting its UK base and smaller funding scale relative to US peers.
The company faces an increasingly crowded AI safety field, with competitors like Anthropic, DeepMind, and OpenAI also investing heavily in alignment research. Conjecture's approach—building AI from modular components rather than scaling monolithic models—remains unproven at commercial scale, and the company has not demonstrated a working product in production environments. As of June 2026, Conjecture appears to be in an early research and development phase, with limited public traction.
Products by Conjecture
Who buys this
- Enterprise AI teams seeking interpretable and controllable AI systems for complex workflows
- Government and defense agencies focused on AI safety and alignment
- Research institutions studying mechanistic interpretability and AI alignment
- Companies automating multi-step business processes that require high reliability
Strengths and what to watch
Strengths
- Founding team includes researchers from EleutherAI, a respected open-source AI research collective, giving the company credibility in the AI safety community
- Cognitive Emulation approach offers a differentiated technical path that avoids scaling large language models, potentially reducing risks like hallucination and unintended behavior
- Raised approximately $150 million from top-tier investors including Index Ventures and Factorial Funds, providing a multi-year runway for research and development
Watch for
- No publicly disclosed customers, revenue, or working product in production, raising questions about commercial viability and market traction
- Competition from well-funded AI safety teams at Anthropic, DeepMind, and OpenAI, which have larger resources and established research output
- Small team size and limited hiring activity (only one open role as of June 2026) may constrain development speed and ability to deliver on ambitious technical goals
Key Information
- Industry
- Commercial AI Research
- Founded
- 2022
- Headquarters
- London, UK
Frequently Asked Questions
What is Conjecture and what does it do?
Conjecture is a London-based AI safety startup founded in 2022 that builds AI systems using a modular architecture called Cognitive Emulation. This approach mimics human expert reasoning to improve reliability, interpretability, and control, addressing issues like hallucination and incoherence.
What is Cognitive Emulation in AI?
Cognitive Emulation is Conjecture's proprietary AI architecture that decomposes complex tasks into reusable, human-expert-guided pipelines. It aims to make AI systems more interpretable and controllable by following human-like reasoning processes, offering an alternative to scaling large language models.
How much funding has Conjecture raised and who invested?
Conjecture has raised approximately $150 million in funding from investors including Index Ventures, Factorial Funds, and angel investor Jaan Tallinn. This funding provides a multi-year runway for research and development, as reported in 2023.
Does Conjecture have any customers or revenue?
As of June 2026, Conjecture has not disclosed any customers or revenue. The company appears to be in an early research and development phase with no working product in production, raising questions about commercial viability.
How does Conjecture differ from Anthropic and DeepMind?
Conjecture focuses on a modular architecture called Cognitive Emulation rather than scaling large language models, which it argues amplifies safety risks. However, it faces competition from well-funded AI safety teams at Anthropic, DeepMind, and OpenAI with larger resources.
What are the main risks or challenges for Conjecture?
Key challenges include no publicly disclosed customers or revenue, competition from larger AI safety teams at Anthropic and DeepMind, and a small team with limited hiring. Its approach remains unproven at commercial scale as of mid-2026.
Sources
- www.conjecture.dev — Company description, product details, Cognitive Emulation architecture, job posting, and information hazard policy
- techcrunch.com — Confirms Conjecture is not listed among US AI startups raising $100M+ in 2025, indicating its smaller funding scale and UK base
- s206.q4cdn.com — Provides context on AI industry financials (Alphabet Q1 2026) but does not mention Conjecture; used to show competitive landscape
- investor.figma.com — Provides context on AI adoption in enterprise (Figma Q1 2026) but does not mention Conjecture; used to show market trends