Aayush Bhan
Agentic AI Engineer
AI Safety & 
Control.
Keep people able to stop, bound, and correct what an AI system does on its own.
AI Safety & Control.
Prevent irreversible AI autonomous actions
Tool-using AI agents often fail to recognize when they should stop. Under ambiguity, conflicting constraints, or tool failures, they may take irreversible actions before identifying the risk, and a later refusal cannot undo the harm.
Enable reliable human override
A stop or override fails when the system is busy, degraded, or already mid-action.
Detect when human intervention is needed
The system keeps going through cases a person should have reviewed.
Prevent AI from bypassing human approval
The agent finds a path that skips the approval a person was supposed to give.
Define safe operating boundaries
The system has no clear limit on what it may do without a person.
Limit autonomous decision-making
Autonomy covers decisions that should stay with a person.
Prevent cascading autonomous actions
One action triggers a chain the operator never approved.
Safely deploy long-running agents
An agent that runs for hours can drift from the task it was given.
Sandbox AI systems
The system can reach tools and data outside the environment it was given.
Restrict system capabilities
The deployed system can do more than the job requires.
Prevent escape from defined environments
The system leaves the network, account, or tool set it was placed in.
Safely isolate malfunctioning systems
A failing system stays connected and keeps acting on live work.
Prevent harmful AI actions
The system can take an action that injures people, property, or rights.
Detect dangerous behaviour
Harmful patterns in outputs or tool use appear and nobody is alerted.
Prevent foreseeable misuse
Ordinary use can turn the system toward a harm the builder already knew about.
Reduce severity of AI failures
When the system fails, the blast radius is the whole task instead of a contained miss.
Prevent unintended optimization
The system pursues a metric that was never the outcome people wanted.
Detect reward hacking
The system scores well by gaming the measure instead of doing the job.
Prevent specification gaming
A literal reading of the instruction produces a result nobody asked for.
Detect divergence between intended and actual behaviour
What the system does no longer matches the behaviour it was built for.
Pick a challenge
Build a solution
Win prizes
Showcase to experts
Pick a challenge
Build a solution
Win prizes
Showcase to experts
Pick a challenge
Build a solution
Win prizes
Showcase to experts
Build with purpose.
Challenges are aligned with the United Nations and USAID Sustainable Development Goals. You're not just learning, you're working on problems that reflect real-world needs.
Meet your mentors
Get guidance from industry professionals.
Gio Masawi
Global Solutions Manager Gemini Enterprise, Agentic AI
Karina Assini Andreatta
Senior Data Scientist, AI Engineer
Casius Sibanda Lee
AI Developer Advocate
Geoffrey Rolin Jacquemyns
Claude Ambassador
Anubhav Pandey
AI Product Manager, Forward Deployed Engineer
Gain invaluable insights and support from seasoned experts who are passionate about driving positive change. Our mentors bring years of experience in tackling complex challenges, ready to share their knowledge and help you make a meaningful impact.
Hop on a team
Collaborate with like-minded.
Choose your challenges and connect with like-minded peers. Work together to tackle crucial social, environmental, and governance challenges with expert guidance.
Tobias
Maria
Ariya
Eugene
Jury Panel
Put your solution in front of hiring managers.
Receive valuable feedback and insights from hiring managers and industry experts.
Get recognised
by the community.
Winner
Finalist
Finalist
Tech partner
The winner and two finalists will receive Exa.ai credits and cash prizes (TBD), exposure through Exa’s developer community, and technical support to help develop their projects further.
Nominees
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
- Shortlisted
FAQ
Learn more about the challenge
Who can participate?
Anyone can join, no matter your background or experience level. Whether you're a student, recent graduate, or transitioning careers, this platform is designed to help you build real-world skills. We welcome diverse perspectives, including scientists, designers, engineers, analysts, marketers, educators, and more.
Is it free?
Yes, participation is completely free.
How to get started?
- Create an account.
- Choose a challenge you would like to work on.
- Follow guides to solve the problem (solo or in a team).
- Submit your solution.
- Receive feedback from experts and see results at the start of each month.
Are there any deadlines?
You can submit your case at any point. Challenge results are announced in the first few days of each month on the site and in community updates.
Can I participate as part of a team?
Yes, participants can submit the case as a team. There is no maximum number of team members, but we recommend up to five members.
How to obtain assistance from mentors?
Mentor support is optional. You can connect with mentors through the platform to get guidance or feedback - but you can also complete challenges independently.
Can participants apply multiple times?
Yes, participants can submit multiple cases, either solo or as part of a team.
Join our community
today.
Join our challenge and turn your ideas into impactful solutions! Collaborate with experts, develop crucial skills, and tackle real-world problems. Your creativity could be the key to driving sustainable change.