About the week.
AI safety belongs in the public.
Opening
One week, across disciplines.
AI Safety Week is an annual convening in Toronto for the people dedicated to making AI safe: how it is built, how it is governed, and how it is understood. Researchers, policymakers, engineers, journalists, civil society, industry, and members of the public come together through talks, workshops, demonstrations and working sessions, and the conversation in between.
The first edition will run from 9 to 13 November 2026, built around The Hinton Lectures.
Purpose
Why the week exists.
Canada has significant AI expertise across research institutes, universities, nonprofits, think tanks, government, industry and independent organizations. On a global stage, Canada is also well placed to bring other middle powers together to help lead this conversation, rather than just respond to it.
AI Safety Week’s efforts are focused on creating a recurring place for this ecosystem to connect, where participants leave with a clearer picture of the field, a wider circle of people to call on, and concrete ideas worth pursuing. Over time, that work strengthens Canada’s AI safety community from within, while giving people around the world a steady way to engage with the work happening here.
Themes
Three questions the week keeps returning to.
Each day of the week takes one of these as its theme.
EVIDENCE
How do we know what an AI system is actually doing? Focuses on the research, evaluations, methods and evidence used to understand AI capabilities, behaviour and risk. Expect technical presentations, research demonstrations and conversations about what current approaches can tell us, where their limits are, and what still needs to be understood.
OVERSIGHT
Who gets to check, and what happens when they find something? Looks at the institutions, tools and rules involved in examining AI systems and acting on what is found. The day connects technical questions with governance, accountability, verification, policy and implementation.
COORDINATION
What do we need to do next, and what requires working together? Focuses on the areas where organizations, disciplines, sectors and countries can make more progress by working more closely together. The focus is on turning the week’s ideas, conversations and new relationships into clearer next steps.
Principles
Four principles
that govern the week.
01
Scientific, not speculative.
The Foundation's mission is to increase awareness of AI's catastrophic risks in a scientific and solutions-oriented manner. That standard governs the week. Claims are sourced, uncertainty is stated plainly and no one is asked to accept a conclusion on authority alone.
02
Solutions-oriented by design.
We take the risks of advanced AI seriously. We also hold that catastrophe is not settled — research, governance and public understanding can change the odds. Every session is expected to end somewhere useful, not merely somewhere alarming.
03
Open where it can be, closed where it must be.
Public lectures are ticketed, with rates for seniors and volunteers. Working sessions are closed so participants can speak plainly. The editorial strand exists for journalists who need to understand the field well enough to report on it accurately.
04
Independent of the labs.
AI Safety Week does not accept sponsorship from frontier AI companies or platform-holders, and the Foundation is independent of government. This is a hard constraint, not a preference. The week is funded by individuals, charitable partners and organizations with no stake in the outcome.
Program
How the week is structured.
Around twenty-eight sessions across five days, in three formats.
Public Lectures
About eight sessions · Ticketed
Evening lectures for general audiences, with no technical background assumed. Each pairs a prepared talk with an extended discussion; questions come from the theatre and from a curated remote queue. The Hinton Lectures — three consecutive evenings convened by Professor Geoffrey Hinton — are the centrepiece. Other lectures take up forecasting, governance and the relationship between capability and risk. Rates for seniors and volunteers.
Working Sessions
About 14 sessions · By application
Half-day working groups held under Chatham House rule, so participants can speak plainly and go on the record only where they choose to. They are built for researchers, senior policymakers and practitioners working directly on AI safety problems. The programme is curated rather than self-selected; apply if the work is yours.
Editorial Events
About six events · Mixed access
Salons, readings and small dinners for journalists, science writers and communicators. The editorial strand connects technical rigour to public narrative, giving writers the vocabulary and context to cover the field accurately. It sits alongside the Foundation's CJF-Hinton Award for excellence in AI safety reporting. Some events are open, others are for accredited press.
The full program will be published by October 2026. Room capacity is limited and the Hinton Lectures are seperately ticketed, so registering is the reliable way to hear when tickets open and reserve your place.
Organiser
Organized by the
AI Safety Foundation.
The AI Safety Foundation is a Canadian registered charity (No. 78867 2822 RR0001) founded in 2024. Its mission is to increase awareness of AI's catastrophic risks in a scientific and solutions-oriented manner, and it works through three pillars: the Hinton Lectures, the CJF-Hinton Award for excellence in AI safety reporting, and AI Safety Week.
The Foundation does not accept sponsorship from frontier AI companies — a hard constraint, not a preference — and is independent of government. It is supported by individuals, corporations and charitable partners. What sets it apart is where it puts its attention: on public understanding rather than research or policy alone, on the conviction that a well-informed public is itself a safety mechanism.