To be decided
2nd place
data leakage& safety
challengeShare a solution that stops an assistant or agent from exposing private data.
Partner prizes being added
Fast-track interviews, fellowships, cash and more

Assistants and agents carry private text into places the user did not choose. A pasted document, a retrieved email, or a tool result can reappear in an answer, a log, or the next tool call, and a later refusal cannot pull that data back.
Nov 15 26’
Challenge start
Dec 15 26’
Deadline
Meet your jury





The jury will be announced soon.
Top solutions
Keeps private data inside the boundary the user intended.
Project name
Author’s names
1st
Project name
Author’s names
2nd
Project name
Author’s names
3rd
Project name
Author’s names
Finalist
Project name
Author’s names
Finalist
Project name
Author’s names
Finalist
Project name
Author’s names
Finalist
Project name
Author’s names
Finalist
Project name
Author’s names
Finalist
Project name
Author’s names
Finalist
Retrieves only the sensitive data needed for the task.
Keeps secrets out of answers and tool arguments.
Prevents private data from resurfacing in later turns or stored logs.
Protects sensitive data without unnecessarily blocking legitimate work.
Winners will be featured on Drawerrr’s Winners page.
The Top 10 projects will be recognised across award categories.
Top solutions among
To be decided
2nd place
To be decided
1st place
To be decided
3rd place
Extra awards from our partners: fast-track interviews, fellowships, cash and more. Exact categories and prizes will be announced soon.
Nov 15 26’
Challenge start
Dec 15 26’
Deadline
How to succeeded

Day 1
Pick an assistant or agent you can run, and name one place private text can leave: a reply, a tool call, a log, or a later turn.
Tip:
Start with a pasted note, a retrieved email, or a secret in a tool result.
Learn more:
Day 2
Recreate it with fake personal data in an inbox, document, or tool you control. Do not use real customer data.
Tip:
A test mailbox with dummy names, emails, and card fragments is enough.
Learn more:


Day 3
Stop the private text from leaving. Redact it before the model sees it, block it in the reply, or keep it out of logs and the next tool call.
Tip:
Check the answer, the tool arguments, and the log. A leak in any one of them counts.
Learn more:
Day 4
Try a secret in the prompt, in retrieved text, in a tool result, and in a follow-up turn. Note what still appears.
Tip:
Also try a normal request that should still be answered, so the safeguard is not only a blanket block.
Learn more:


Day 5
Share the dummy data, the check, and a short record of what leaked and what did not.
Tip:
Publish the test set and the expected redactions so someone else can run them.
Learn more:
FAQ
Every submission is evaluated against the following criteria:
| Criterion | Weight | What judges evaluate |
|---|---|---|
| Safety effectiveness | 35% | Does the safeguard reliably prevent the targeted unauthorized or harmful action? |
| Testing and robustness | 25% | Has the solution been tested against realistic failure scenarios and bypass attempts? |
| Real-world relevance | 20% | Is the problem meaningful, and does the proposed safeguard address it? |
| Reproducibility and reuse | 15% | Can another team understand, test, or adapt the solution? |
| Clarity of communication | 5% | Are the implementation, evidence, and limitations clearly presented? |
| Total | 100% | — |
Our mission is to accelerate practical solutions to real-world AI safety challenges by connecting people, ideas, and expertise to build AI systems that are safer, more reliable, and easier to govern. Drawerrr’s AI safety challenge is informed by leading institutions in AI safety, standards, and public policy.