Fundamentals of Technical AI Safety · Next cohort

Technical AI Safety

A free, six-week online AI safety course from AI Safety India that teaches the AI safety fundamentals, from alignment to interpretability. AI capabilities are advancing faster than our ability to make them safe.

Free · Basic Python · Applications open soon · Starts with each new cohort

Apply nowSee the six weeks
  • 6 weeksLive online, with a facilitator
  • 5 to 7 hoursA week, alongside work
  • 50 peoplePer cohort. Limited seats.
  • GlobalOpen to anyone, anywhere
Technical AI Safety Apply

Who it’s for

  • Researcher · Student

    You know how to learn. You want to point it at something that matters.

    For students and researchers in STEM or the social sciences.

  • Engineer · Practitioner

    You build AI systems. You want to understand what makes them fail.

    For software engineers, ML engineers and data scientists.

  • Curious · Committed

    You are not sure where you fit yet. You want to find out.

    No AI safety background needed. Basic Python is enough.

Live, in small pods, with people who take it seriously.

Attendees in discussion at an AI Safety India meetup in Bengaluru.

Fundamentals of Technical AI Safety, Next cohort

Six weeks, six labs

  1. Week 01The technical challenge with AIMaking AI go wellLab LLM safety and APIs
  2. Week 02Training safer modelsReinforcement learning and RLHFLab RL alignment
  3. Week 03Detecting dangerEvals, red teaming and safeguardsLab Model testing
  4. Week 04Understanding AIMechanistic interpretabilityLab TransformerLens
  5. Week 05Minimising harmRobustness, alignment and jailbreaksLab Adversarial testing
  6. Week 06Start contributingOpen research problems and safety career pathwaysLab Research design
Anusha presenting her research at an AI Safety India meetup in Bengaluru.

A week, in three steps

  1. 01ReadOne reading sets up the question.
  2. 02DiscussTwo hours live, in a pod of 6 to 8.
  3. 03LabHands-on, so the idea becomes yours.

In their own words

From people in Cohort 1.

  • “The facilitation model, not lecture-based, built real critical thinking rather than surface familiarity.”

    NehaPM/BA · UNESCO Women for Ethical AI, South Asia

  • “I was exploring AI Safety on my own. It was scattered. The cohort fixed that: structure, consistency, and people who actually took it seriously.”

    PrasannaNow at an impact-aligned startup

What you leave with

Certificate of completion

Verifiable online

The six labs

  1. week 01labLLM safety and APIs
  2. week 02labRL alignment
  3. week 03labModel testing
  4. week 04labTransformerLens
  5. week 05labAdversarial testing
  6. week 06labResearch design
  7. donesix labs, one research design

Questions

Anything else? Write to contact@aisafetyindia.org.

What is Fundamentals of Technical AI Safety?
Fundamentals of Technical AI Safety is a free, six-week online course from AI Safety India. Each week you read, discuss for two hours live in a pod of 6 to 8 with a facilitator, and do a hands-on lab on alignment, RLHF, evals, interpretability or robustness.
Do I need prior AI safety knowledge?
No. Basic Python is enough. What matters most is 5 to 7 hours a week and genuine motivation to engage with this hard, important problem.
What does it cost, and when is the next cohort?
The 6-week course is free. Each cohort’s dates show here when applications open.
What will I be able to do after?

Understand alignment failure modes, implement interpretability techniques, evaluate model safety protocols, and analyse the policy implications of technical work.

Graduates go on to roles such as AI safety research fellow, model alignment engineer, evaluation specialist and safety policy analyst.

Is this only for people from India?
No. The cohort is open globally. Cohort 1 had people from nine countries across four continents. The programme is anchored in India’s regional context but open to contributors worldwide.
What happens after the 6 weeks?
You receive a certificate of completion and join the AISIN alumni network. From 2027, high-performing graduates will be able to apply for a 12-week research fellowship with a stipend, producing real published outputs.
Who leads the discussions?
A facilitator leads each pod’s weekly discussion, guiding it rather than lecturing. Want to facilitate a future cohort? See the volunteer facilitator role for Technical AI Safety.
Not sure which track? Compare them
CompareTechnical AI SafetyAI Governance
Who it is forEngineers, ML people, STEM researchers and studentsPolicy, law, social science, economics and industry roles
Background neededBasic PythonNone, no coding
Each week you makeA lab: code, evals, interpretabilityAn exercise: scans, audits, a policy memo
You finish withA research designA real policy memo and your leverage point
ThenResearch fellowship, technical safety trackGovernance and policy research fellowship

See AI Governance

The field needs more people.

Apply now
Researcher · Student

You know how to learn. You want to point it at something that matters.

You are in a STEM or social science field. You have heard of AI safety but the landscape feels scattered. This programme gives you a structured map: the open problems, where your skills fit, and who is working on what.

Engineer · Practitioner

You build AI systems. You want to understand what makes them fail.

You are a software engineer, ML engineer or data scientist. The safety field can feel academic and hard to enter from industry. This is the on-ramp: a working understanding of technical safety problems and people doing this work at labs.

Curious · Committed

You are not sure where you fit yet. You want to find out.

No prior AI safety background required. Basic Python is enough. What matters is 5 to 7 hours a week and genuine motivation to engage with a hard, important problem.

Programmes

Hubs