A nonprofit institute
Trust and safety standards for AI agents.
TSI helps organizations build, assess, and demonstrate trustworthy AI agents. We convene leaders from industry, engineering, and research to define what trustworthy deployment takes in practice, and publish it openly, so agents can take on meaningful work with confidence.
Trusted by leading global organizations
AI Square
OpenGrowth Ventures
Nasiko
Council on Competitiveness
EAF
Morgan Signing House
Behavioural
imodes.ai
Why we exist
Trust is what lets capability scale.
AI agents are ready to take on meaningful work. Strong governance gives organizations the confidence, clarity, and accountability to let them. TSI brings the market together to turn that confidence into usable standards, rigorous evaluation, and shared learning.
Informed deployment
Teams see where an agent is ready, where it needs support, and what evidence stands behind each decision.
Enterprise confidence
Executives, buyers, and risk teams share one clear view of readiness, so capable agents reach production sooner.
Trusted scale
Continuous assessment and shared learning let agents take on more responsibility over time.
No organization builds that confidence alone. That is why we convene.
The Collective Intelligence Trust Stack
Trust is not a policy. It is seven controls.
Seven questions every deployed system must be able to answer. Use them to assess maturity, strengthen readiness, and chart a credible path to trusted scale.
- 01
Human purpose
What human benefit is this system designed to advance?
- 02
Identity and authority
Who is acting, and under whose authority?
- 03
Data and knowledge stewardship
Is the data and knowledge behind it governed?
- 04
Explainable execution
Can a reviewer reconstruct exactly what happened?
- 05
Simulated intelligence
Was it tested against adversarial conditions first?
- 06
Runtime governance and token monitoring
Is it monitored and governed while it runs?
- 07
Ecosystem sharing and learning
Are lessons and incidents shared across the field?
How the Trust Stack assessment metrics are developed
- Relevant best practices for AI governance and risk management
- International trust, safety, security, privacy, and governance practices
- Ongoing input from TSI’s working group of international advisors
The Trust Stack is decision support for organizations deploying AI agents. It complements, and does not replace, legal, regulatory, or sector-specific certification requirements.
The TSI Scoring Tool
Know whether your AI agents are ready for real work.
The TSI Scoring Tool independently assesses AI agents against global standards and the TSI Trust Stack, covering the controls that matter in production, from identity, authority, and data and knowledge stewardship to explainability, adversarial testing, runtime governance and token monitoring, and ecosystem sharing.
- Assess readiness
- Identify priority improvements
- Build a clear assurance record
- Create a defensible path to deployment
- Identity & authorityAssess
- Data and knowledge stewardshipAssess
- Explainable executionAssess
- Runtime governance and token monitoringAssess
- Ecosystem sharing and learningAssess
A defensible score, informed by established trust and safety standards, with the actions needed before deployment.
TSI Trust Tags · Part of the TSI Score
A clear rating for every assessed agent.
Every TSI Score, independently assessed against global standards and the TSI Trust Stack, maps to a rating from A1 to A5, so readiness is easy to read at a glance and easy to improve over time.
- A1Foundational
Early-stage controls are defined; further assurance is needed before meaningful autonomy.
- A2Developing
Core controls are emerging and readiness is improving.
- A3Assured
The agent demonstrates a credible trust and governance baseline for its intended context.
- A4Advanced
Strong, repeatable controls with clear evidence, oversight, and operational maturity.
- A5Trusted at Scale
Leading assurance posture for high-confidence deployment, continuous monitoring, and ecosystem sharing and learning.
TSI Trust TagExample for illustration. Final Trust Tag design and use are set by TSI programme requirements.
The TSI Trust Tag helps organizations communicate that an AI agent has been independently assessed for trustworthy deployment.
Why a Trust Tag matters
- Makes agent readiness easier to understand across technical and non-technical stakeholders
- Creates a common language for procurement, governance, risk, and executive teams
- Signals a commitment to transparent, responsible deployment
- Gives solution providers a clearer path to demonstrate assurance
A completed assessment can lead to the appropriate TSI rating and, where qualification requirements are met, TSI certification and use of the TSI Trust Tag. Ratings, certification, and Trust Tag use are subject to TSI’s final methodology, assessment evidence, and programme requirements. TSI assessment complements, and does not replace, legal, regulatory, or sector-specific certification requirements.
Community · AI Debrief Room
The AI Debrief Room is where the market shares what it takes to deploy AI agents with confidence.
A curated TSI knowledge and discussion space for enterprise leaders, advisors, researchers, practitioners, and AI solution providers. Insights, research, recordings, discussion, and downloadable TSI Market Intelligence, in one room.
Featured Insights
TSI Market Briefs
Articles
Research Reports
Advisor Perspectives
Industry Trends
Webinars & Event Recordings
Discussion Topics
Downloadable Resources
Join the AI Debrief Room
Get TSI Market Briefs, research, recordings, and invitations to roundtables.
International working group
Standards shaped by the people who use them.
TSI standards and assessment methodology are shaped by an international working group of advisors, informed by real enterprise deployment needs, and continuously improved as the market matures.
- Enterprise leaders
- AI engineers
- Security and risk leaders
- Researchers and academics
- Policy and legal advisors
- AI solution providers
- 01
Convene
The people who answer for agents, build them, and study them, in one room.
- 02
Test
Ideas are tested against real systems and real enterprise deployment needs.
- 03
Agree
Specific enough to be useful, open enough to be challenged.
- 04
Publish
Free to read, free to use, free to build on.
- 05
Assess
Published standards become the basis of the TSI Score and Trust Tag.
Working groups
Join the working group.
Groups are forming around each control. Chairs are open.
TSI × AI Square · Reasoning Markup Language
TSI defines the trust and assurance framework. AI Square operationalizes trusted agent decisions through Reasoning Markup Language, making claims, assumptions, evidence, policy logic, and human corrections inspectable and reusable.
- TSIGlobal standards and the TSI Trust Stack
- BridgeRML evidence and reasoning
- AI SquareAI Square operationalization
- TSITSI Score / Trust Tag
- TSIAn independent institute and standards authority.
- AI SquareA separate enterprise technology platform.
- RMLA practical bridge between TSI’s trust framework and day-to-day agent operations.
Trust and Safety Institute
- Independent institute and standards authority
- Defines the trust and assurance framework
- Issues the TSI Score, ratings, and Trust Tag
- Separate enterprise technology platform
- Operationalizes trusted agent decisions through Reasoning Markup Language
Get involved