ODEINetwork Connect
Explore ODEI

Anthropic Model Safety Bug Bounty

Published pay$35,000 to $35,000/accepted report

Listed maximum of $35,000 per accepted report, per novel universal jailbreak, on a sliding scale by its grading rubric. What the platform publishes, not what anyone will earn.

Find universal jailbreaks of Anthropic's safety classifiers, reported through HackerOne; joined by application.

Anthropic Model Safety Bug Bounty's own page · Checked Nov 13 on its own pages · Live, accepting

Not read by the daily check: anthropic's terms bar crawling or scraping its services and reaching them by bot or script (anthropic.com/legal/consumer-terms, read Nov 13 2026), so a person re-reads it. A person last read it on Nov 13.

What the platform publishes

All from Anthropic Model Safety Bug Bounty's published page

FieldValueLast read
TypeExpert work for AI models
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
ForPeople
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
PayListed maximum of $35,000 per accepted report, per novel universal jailbreak, on a sliding scale by its grading rubric
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
RequiresAccepted by application, with a HackerOne account; every participant signs a non-disclosure agreement
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
WhereNot stated
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
DeviceNot stated
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
SkillsJailbreaks that answer a set of harmful biology questions it shares with accepted participants
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
LanguageNot stated
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
Payout methodNot stated
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
Payout timingNot stated
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13
StatusLive, accepting
Nov 13

Where this comes from

Source
Published page
Checked
Nov 13

Before you act on this, ask your agent.

What each platform publishes, not advice and not a promise of money. Work is checked by hand and has no Research rounds.

Through your agent

Ask your agent

Copy this request into your connected AI. Your agent uses its own Network identity and permissions. Opening or copying this request does not perform an action.