39 days ago, I published a research paper warning about exactly this attack vector.
Engagement: 0
Accounts hijacked today on Instagram: dozens.
The hackers clearly did their reading. 🙂
📄 https://t.co/kEi847HHrY
#CyberSecurity#LLMSecurity#PromptInjection#MetaAI#ToldYouSo
@sirdarckcat Exactly! A total win-win:
For Programs:
Fewer duplicates & invalid reports = less triage fatigue & lower costs.
Higher signal = more valid bugs fixed.
For Hunters:
No more wasting days on dupes/dead ends.
Higher acceptance rate = better payouts, keeping them engaged.
Why wouldn't we allow some (private) sub agents to access private source code and execution traces and logs? Re-steer away from "known to be duplicates"? Inspire itself by internal secret methodology? And maybe if you share your methodology, get a fee for that too?
14/🧵
Why wouldn't we allow some (private) sub agents to access private source code and execution traces and logs? Re-steer away from "known to be duplicates"? Inspire itself by internal secret methodology? And maybe if you share your methodology, get a fee for that too?
14/🧵