AIprompt injectionsafetychallenge

Share a solution that stops untrusted text from making an agent call a tool or ignore the user.

Partner prizes being added

Fast-track interviews, fellowships, cash and more

OWASP lists prompt injection as LLM01:2025, the first risk in its Top 10 for LLM applications.

* Prompt Injection ↗

The agent treats outside text as an instruction. A webpage, an email, or a tool response can tell it to send data, change a record, or skip the user’s request, and the call looks like ordinary work.

Nov 15 26’

Challenge start

Dec 15 26’

Deadline

Meet your jury

Your work will be evaluated
by industry experts.

The jury will be announced soon.

Top solutions

Winners

Best overall prompt-injection defence

Stops untrusted content from changing the user’s request or causing an unauthorized action.

  • 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

Best in authorizing tool calls

Checks each proposed call against the user’s intent before it runs.

Best in separating instructions from content

Treats webpages, emails, files, and tool results as untrusted data, not as orders.

Best in resisting adaptive bypasses

Keeps working across varied attacks, obfuscation, and malicious tool output.

Best in making security decisions visible

Shows the injected instruction, the decision, and any blocked action.

Winners will be featured on Drawerrr’s Winners page.
The Top 10 projects will be recognised across award categories.

Top solutions among

Partner categories

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

Master the Challenge in 5 days

Day 1

Identify high-risk scenarios

Pick a tool-using agent you can run, and name one action untrusted text could trigger: send data, change a record, or ignore the user.

Tip:

Start with a page, email, or file the agent will read, with a hidden instruction to call a tool.

Learn more:

Day 2

Set up a test environment

Rebuild that path in a sandbox. Give the agent one harmless tool and a fixture that contains the injected instruction.

Tip:

Use a local page or email and a tool that only writes to a log.

Learn more:

Day 3

Choose a safeguard stack

Put a check between the untrusted text and the tool call. Separate instructions from content, require approval, or block the call.

Tip:

Treat page and tool text as data, and let only the user instruction authorize a tool.

Learn more:

Day 4

Document failure states

Try hidden instructions, conflicting requests, malicious tool output, and a call the user actually wanted.

Tip:

Record whether the tool ran, whether a legitimate call was blocked, and what the user saw.

Learn more:

Day 5

Make it reproducible

Package the fixture, the check, and the results so someone else can replay the injection and the block.

Tip:

A README with the fixture and the expected tool log is enough when the agent itself cannot be shared.

Learn more:

FAQ

Learn more about the challenge

1. What is the Prompt Injection & Agent Security challenge?

This challenge asks participants to stop untrusted text in webpages, emails, files, or tool responses from overriding the user or triggering an unauthorized tool call. Share a solution that stops untrusted text from making an agent call a tool or ignore the user. Potential areas include: • Separating trusted instructions from untrusted content • Tool authorization and human approval • Policy enforcement before execution • Detection of malicious content in tool output • Prompt-injection tests with legitimate-action controls

2. Who can participate in the challenge?

The challenge welcomes students, career switchers, working professionals, independent researchers, engineers, and others interested in practical AI safety. You do not need to be affiliated with a university or school. The affiliation field is optional. If your submission involves an employer’s intellectual property or confidential information, make sure you have permission to share it.

3. Can I participate individually, or do I need a team?

You may participate individually or as a team. An individual submission can suit a focused prototype, evaluation, or case study. A team may help when the work spans engineering, security, testing, hardware, or user experience. There is no maximum team size; we recommend up to five members.

4. What can I submit?

You can submit an existing project, an improvement to an existing solution, or a new approach that clearly addresses prompt injection & agent security. Example approaches: • Separating trusted instructions from untrusted content • Tool authorization and human approval • Policy enforcement before execution • Detection of malicious content in tool output • Prompt-injection tests with legitimate-action controls You may submit a reusable agent, Skill, Bot, plugin, MCP, custom build, test harness, implementation, or article/case study.

5. How is it judged?

Every submission is evaluated against the following criteria:

CriterionWeightWhat judges evaluate
Safety effectiveness35%Does the safeguard reliably prevent the targeted unauthorized or harmful action?
Testing and robustness25%Has the solution been tested against realistic failure scenarios and bypass attempts?
Real-world relevance20%Is the problem meaningful, and does the proposed safeguard address it?
Reproducibility and reuse15%Can another team understand, test, or adapt the solution?
Clarity of communication5%Are the implementation, evidence, and limitations clearly presented?
Total100%—

6. What should I include in my submission?

Strong submissions define a clear safety problem, explain how the approach works, provide evidence, test relevant failure scenarios, and document limitations honestly. Depending on maturity, include a project overview, technical explanation, demonstration, tests and results, reproducibility instructions, and known limitations. Early-stage ideas and prototypes are welcome when you clearly separate what is implemented, tested, and still theoretical.

7. Do I need to open-source my project?

No. You may submit a plugin, connector, app, or an article or case study describing work in a closed system. If you share code or materials, remove credentials, personal data, and security-sensitive details. For a closed system, provide enough evidence and instructions for judges to understand and evaluate the approach.

8. When is it due?

Submit by Dec 15, 2026. Winners are announced Dec 31, 2026.

9. What prizes or recognition are available?

The strongest projects will be featured on Drawerrr’s Winners page, and the Top 10 projects will be recognised across award categories. Extra awards from our partners: fast-track interviews, fellowships, cash and more. Exact categories and prizes will be announced soon.

10. Will submissions be reviewed by experts?

Yes. Submissions are evaluated by a jury of industry experts using the published criteria. Jury members will be added to the challenge page as they confirm.

11. Can I update my submission after submitting it?

Yes. You can submit again with updated materials before the deadline. Each submission creates a new record, so use the latest submission for your final project link and details.

12. Who owns the intellectual property in my submission?

You retain ownership of your work. By submitting, you grant Drawerrr permission to display and evaluate the submission for challenge operations, mentoring, and jury review. Do not include confidential material you are not authorized to share, and confirm you have rights to employer, collaborator, or third-party content.

Subscribe to challenge results

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.

European AI OfficeAISINIST