AI Security Institute
The AI Security Institute is a UK government research directorate within DSIT that conducts pre-deployment safety and security evaluations of frontier AI models, develops open-source evaluation tools and benchmarks (Inspect, ControlArena, RepliBench), and administers grant programmes (the Alignment Project with £27M+) for external AI alignment research serving policymakers, frontier AI developers, and international partner institutes.
- Company typePrivate
- Founded2023
- HeadquartersLondon, United Kingdom
- Headcount51–100
- GTM typeB2B
- OfferingServices
What AI Security Institute does
The AI Security Institute (AISI) is a UK government research organization established in November 2023 at the Bletchley Park AI Safety Summit as the Frontier AI Taskforce, later rebranded to the UK AI Safety Institute and then to the AI Security Institute in February 2025 to reflect a strategic narrowing toward security-focused risks. It operates as a directorate within the Department for Science, Innovation and Technology (DSIT) and is described as the world's largest government team dedicated to AI safety and security research, with over 100 technical staff. AISI conducts scientific evaluations of frontier AI models from labs including Anthropic, OpenAI, and Google DeepMind across cyber offense, biology, chemistry, autonomy, and societal-resilience domains, and translates findings into policy guidance for UK and international governments.
Its portfolio includes open-source evaluation tooling (the Inspect framework, ControlArena, Inspect Scout), specialised benchmarks (SandboxEscapeBench, RepliBench, AgentHarm), and grant programmes such as the Alignment Project (over £27 million committed across 60 initial projects, funded by OpenAI, Microsoft, Anthropic, and AWS), the Challenge Fund (up to £200,000 per project), and the Systemic Safety Grants Programme (20 projects). AISI maintains research partnerships across the UK (NCSC, Alan Turing Institute, Defence Science and Technology Laboratory) and abroad (US AI Safety Institute/CAISI, Australian AI Safety Institute, France Inria, MLCommons, Linux Foundation Appia Foundation), and coordinates the International Network for Advanced AI Measurement, Evaluation and Science.
AISI's revenue model is non-commercial: it is funded through UK government departmental budgets and industry contributions routed through grant mechanisms, and does not charge fees for services. Findings, tools, and evaluations are disseminated through open publication on aisi.gov.uk, open-source repositories, academic venues (including ICLR 2025), and direct sharing with policymakers, frontier AI developers, and international government partners. Its customers in a functional sense are governments/policymakers, frontier AI developers, academic researchers, international AI safety institutes, and UK critical-infrastructure operators, but no commercial sales motion is operated.
AI Security Institute firmographics
Firmographics- Name
- AI Security Institute
- Legal name
- The AI Security Institute (AISI)
- Website
- https://aisi.gov.uk
- Company type
- Private
- Founded year
- 2023
- Operating status
- Operating
- Headcount range
- 51–100 employees
- Short description
- The AI Security Institute is a UK government research directorate within DSIT that conducts pre-deployment safety and security evaluations of frontier AI models, develops open-source evaluation tools and benchmarks (Inspect, ControlArena, RepliBench), and administers grant programmes (the Alignment Project with £27M+) for external AI alignment research serving policymakers, frontier AI developers, and international partner institutes.
- Ownership category
- akta.pro rank
AI Security Institute industry classification
Industry- Product category
- AI Safety Research & Government Evaluation Services
- NAICS
- Public Administration (92), National Security and International Affairs (928)
- SIC
- Services-Computer Integrated Systems Design (7373)
- akta.pro primary industry
- AI Governance, Risk & Compliance (GRC) Platforms (HDAAAMAA)
- akta.pro secondary industries
- Safety & Alignment Evaluation (red-teaming, harmful capability testing) (HDAAAMAL), AI Observability, Monitoring & Evaluation Platforms (Drift, Quality, Safety) (HDAEANAF), Responsible AI, AI Governance & Compliance Services (BPAEAHAJ), Audit, Explainability & Accountability Tooling (traceability, reporting) (HDAAAKAL)
Keywords
Where AI Security Institute is headquartered
LocationHeadquarters
- HQ city
- London
- HQ country
- United Kingdom
- HQ region
- Europe
Offices1 record
Markets served
AI Security Institute business model
Business model- GTM type
- B2B
- Offering type
- Services
- Cost components
- Personnel, Technology or R&D, Operations, Others
Revenue model
- Government Funding: AISI is a government research organization within the Department of Science, Innovation and Technology (DSIT), funded by UK government. It receives contributions from industry partners including OpenAI, Microsoft, Anthropic, AWS for specific research initiatives like the Alignment Project.
Go-to-market motion1 record
Distribution channels3 records
Marketing channels5 records
AI Security Institute product offering
Product offeringCore offering
AISI is a UK government research organization that conducts scientific evaluations of frontier AI models to understand their capabilities and risks across cyber, biology, chemistry, autonomy, and societal resilience domains. It develops open-source evaluation frameworks and benchmarks (Inspect, ControlArena, AgentHarm, RepliBench, SandboxEscapeBench), publishes research findings, and runs grant programs (Alignment Project, Challenge Fund, Systemic Safety Grants) to advance AI safety research and inform evidence-based AI governance policy.
Product overview
The AI Security Institute (AISI) is a UK government research organization within the Department of Science, Innovation and Technology, established in November 2023. Rather than a commercial product company, AISI operates as the world's largest government team dedicated to AI safety and security research. Their portfolio consists of open-source evaluation tools (Inspect framework), research benchmarks (ControlArena, SandboxEscapeBench, AgentHarm, RepliBench), grant programs (Alignment Project with £27m funding, Challenge Fund up to £200k per project, Systemic Safety Grants), and research publications including the annual Frontier AI Trends Report. AISI conducts pre-deployment evaluations of frontier AI models from companies like Anthropic and OpenAI, and partners with leading AI developers to advance AI safety science.
Differentiator
Problem solved
Functional benefit
Brands
- The Alignment Project: A global fund supporting AI alignment research with over £27 million in total funding from government, industry, and philanthropic partners.
- Challenge Fund
- Inspect
- ControlArena
Products and services
- Inspect Open-source evaluation framework for testing large language model capabilities, providing facilities for prompt engineering, tool usage, multi-turn dialogue, and model-graded evaluations. Used by AI safety researchers and developers worldwide.
- ControlArena A library for running AI control experiments in realistic mock environments replicating frontier AI developer infrastructure. Tests whether AI systems can subvert control protocols and exfiltrate sensitive data. Used by AI safety researchers.
- SandboxEscapeBench A benchmark evaluating whether AI agents with shell access can escape container sandboxes and retrieve protected files from host systems, testing 18 escape scenarios. Used by AI safety researchers evaluating frontier models.
- AgentHarm A benchmark for measuring LLM agent misuse, published at ICLR 2025, with thousands of GitHub downloads and citations by OpenAI and Anthropic. Evaluates whether leading LLMs refuse or comply with malicious agent requests.
- RepliBench A comprehensive benchmark measuring autonomous replication capabilities in AI systems, detecting emerging replication abilities and providing quantifiable understanding of potential risks across 50 evaluation tasks.
- Frontier AI Trends Report AISI's annual public report on frontier AI capability trends based on evaluations across cyber, chemistry and biology, and autonomy domains since November 2023. Used by governments, policymakers, and AI developers worldwide.
- Alignment Project A global research fund providing over £27 million for AI alignment research across 60 initial projects, supporting independent safety research with contributions from OpenAI, Microsoft, Anthropic, and AWS. For academic researchers and AI safety organizations.
- Challenge Fund
Quantifiable outcome
- First model to complete both full cyberattack simulation environments (Claude Mythos Preview)
- +3 more outcomes
Companies that use AI Security Institute
Customer profileNamed customers5 records
Segments5 records
Ideal customer profiles4 records
AI Security Institute technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability8 records
Feature5 records
AI Security Institute partnerships and signals
Strategic signalPartnerships
17 partnerships are on record, tiered major and minor.
- US Center for AI Standards and Innovation (CAISI)majorOpenAI established testing partnerships with the US Center for AI Standards and Innovation (CAISI) to advance frontier capability assessments and biological-misuse safeguards, alongside the UK AI Security Institute.
- Linux Foundation Appia FoundationminorOpenAI helped found the Appia Foundation, hosted by the Linux Foundation, to develop open, modular specifications that translate international AI standards into practical assessment criteria across the AI value chain.
- Australian AI Safety InstitutemajorUK and Australia signed a Memorandum of Understanding establishing a partnership between UK AI Security Institute and Australian AI Safety Institute. The agreement includes sharing information on frontier AI capabilities, collaborating on AI evaluation best practices, conducting joint research into emerging risks, and facilitating staff exchanges between the two institutes.
- MicrosoftmajorMicrosoft announced new partnerships with CAISI in the US and AISI in the UK to advance AI testing and evaluation science for frontier models. The collaborations focus on adversarial assessments, evaluating safety safeguards, and researching methods to mitigate national security and large-scale public safety risks from advanced AI systems. Microsoft also contributes to global AI safety initiatives including the International Network for AI Measurement, Evaluation and Science and MLCommons.
- ElevenLabsminorPartnership exploring the security and societal implications of voice AI systems.
- International Network for Advanced AI Measurement, Evaluation and SciencemajorAISI serves as Network Coordinator for the International Network for Advanced AI Measurement, Evaluation and Science, bringing together industry, civil society, academia, and governments to develop reliable methods for assessing AI systems.
- Google DeepMindmajorGoogle DeepMind announced an expanded partnership with UK AI Security Institute through a Memorandum of Understanding to focus on foundational AI safety and security research. The collaboration aims to develop monitoring techniques, investigate ethical implications, and evaluate AI's impact on economic systems. Includes joint research areas: monitoring chains-of-thought for alignment, investigating socio-affective alignment, and studying AI's impact on economic activity.
- Linux FoundationminorThe Linux Foundation hosts the Appia Foundation for developing open, modular specifications that translate international AI standards into practical assessment criteria.
- EleutherAIminorCollaboration with EleutherAI and University of Oxford on filtering harmful content from training data to resist adversarial fine-tuning.
- Alan Turing InstituteminorCollaboration on data poisoning research showing that as few as 250 malicious documents can introduce backdoor vulnerabilities in AI models.
- National Cyber Security Centre (NCSC)majorPartnership with NCSC for cyber security research including work on how cyber defenders can use advanced AI capabilities and adapting vulnerability disclosure for AI safeguards.
- MLCommonsminorAISI contributes to MLCommons, which develops testing tools like the AILuminate benchmark for AI safety evaluation.
- Centre for Long-Term ResilienceminorFunded by UK's AI Security Institute for research on AI agent behavior and misalignment.
- Laboratory for AI Security Research (LASR)minorPartnership with LASR for cross-government cyber expertise and AI-driven cyber risk scenario identification.
- Defence Science and Technology LaboratoryminorPartnership with Defence Science and Technology Laboratory for national security community engagement.
- US AI Safety InstitutemajorUK and US AI Safety Institutes signed a landmark agreement to jointly test advanced AI models, share research insights, share model access, and enable expert talent transfers.
- France Inria (National Institute for Research in Digital Science and Technology)minorUK AI Safety Institute partnered with France's Inria to advance AI safety research.
Scale indicators6 records
Recent moves6 records
Expansion highlights6 records
AI Security Institute competitors and assessment
Company assessmentBroad incumbents
- MITRE Corporation: A federally funded R&D center operating AI-related safety and security programs (including ATLAS for adversarial ML). Comparable in the public-interest research and government-federation model, with deeper defense/cyber heritage than AISI.
- RAND Corporation: A policy research organization producing independent analysis on AI safety, national security implications, and frontier technology risks. Overlaps with AISI on policy-facing AI risk research and government advisory work, with a much broader research mandate.
- NIST (National Institute of Standards and Technology): US federal agency that develops technical standards including the AI Risk Management Framework (AI RMF). Comparable in standards-setting role and government-backed evaluation authority, though NIST operates broadly across many technology domains.
Emerging players
- Centre for the Governance of AI (GovAI): A research organization focused on AI governance and policy, often cited as a thought partner to government AI safety efforts. Overlaps with AISI on policy translation work but focuses on governance research rather than empirical model evaluation.
- MLCommons: An open engineering consortium that develops AI benchmarks including AILuminate for AI safety evaluation. AISI contributes as a partner; comparable in building shared evaluation infrastructure used by industry and governments.
Direct peers
- US AI Safety Institute (CAISI): The US counterpart established under the same 2023 Bletchley Park framework, conducting frontier AI evaluations and joint research. AISI has a formal MOU with CAISI for shared model testing, talent transfers, and research collaboration — making it the closest structural peer.
- METR (Model Evaluation and Threat Research): An independent nonprofit that conducts rigorous evaluations of frontier AI capabilities, particularly autonomous task completion. Comparable in research scope and methodology (capability benchmarks, autonomous agent evaluation), though privately funded rather than government-backed.
- Apollo Research: An AI safety research organization specializing in evaluating frontier models for dangerous capabilities, deception, and scheming behavior. Operates in the same evaluation niche as AISI's safety work and frequently collaborates on capability benchmarks.
- Australian AI Safety Institute: Australia's equivalent government body for frontier AI evaluation. The two institutes signed an MOU in May 2026 covering capability information sharing, evaluation methodology, joint research, and staff exchanges — directly comparable mission and operating model.
Regional players
- Alan Turing Institute: The UK's national institute for data science and AI, collaborating with AISI on data poisoning and alignment research. Comparable in UK public-sector AI research positioning, with broader academic rather than safety-specific remit.
Market position
Strengths4 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights6 records
Customer concentration
AI Security Institute social profiles
Digital presenceAI Security Institute financial estimates
Financial estimateRevenue estimate
Valuation estimate
AI Security Institute leadership team
Management profileNumber of profiles
Profiles3 records
AI Security Institute funding detail
Funding detailFunding overview
Funding rounds
Investors
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
AI Security Institute M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about AI Security Institute
What does AI Security Institute do?
AISI is a UK government research organization that conducts scientific evaluations of frontier AI models to understand their capabilities and risks across cyber, biology, chemistry, autonomy, and societal resilience domains. It develops open-source evaluation frameworks and benchmarks (Inspect, ControlArena, AgentHarm, RepliBench, SandboxEscapeBench), publishes research findings, and runs grant programs (Alignment Project, Challenge Fund, Systemic Safety Grants) to advance AI safety research and inform evidence-based AI governance policy.
Is AI Security Institute a public or private company?
AI Security Institute is a private company. It is classified as state government owned and is currently operating.
When was AI Security Institute founded?
AI Security Institute was founded in 2023. It employs 51 to 100 people.
Where is AI Security Institute based?
AI Security Institute is headquartered in London, United Kingdom, in the Europe region.
How does AI Security Institute make money?
One revenue line is on record: government Funding.
Who are AI Security Institute's main competitors?
Broad incumbents on record are MITRE Corporation, RAND Corporation and NIST (National Institute of Standards and Technology). Emerging players are Centre for the Governance of AI (GovAI) and MLCommons. Direct peers are US AI Safety Institute (CAISI), METR (Model Evaluation and Threat Research), Apollo Research and Australian AI Safety Institute. Alan Turing Institute is listed as a regional player.
Does AI Security Institute have an API?
No public API is recorded for AI Security Institute.
What industry is AI Security Institute in?
AI Security Institute's product category is AI Safety Research & Government Evaluation Services. Its primary akta.pro industry code is HDAAAMAA, AI Governance, Risk & Compliance (GRC) Platforms, with a secondary code of HDAAAMAL, Safety & Alignment Evaluation (red-teaming, harmful capability testing). Its NAICS code is 92 and its SIC code is 7373.