MANIFOLD
Will prompt injection be effectively solved by end of 2026?
4
Ṁ1kṀ355
Dec 31
35%
chance

Resolves YES if by December 31, 2026, at least 5 AI safety/security researchers from at least 3 different organizations publicly state that prompt injection has been effectively solved for production systems. At least one organization must be independent (not OpenAI, Anthropic, Google DeepMind, Meta AI). Researchers must have published AI safety/security work or work at recognized AI safety organizations.

  • Update 2026-02-28 (PST) (AI summary of creator comment): The creator has updated the resolution criteria in response to feedback about the title conflicting with the original criteria (which focused on claims rather than actual solving of prompt injection).

Market context
Get
Ṁ1,000
to start trading!
Sort by:

The title says this market is about whether it's solved, but the resolution criteria are about CLAIMS that it has been solved. According to how this is written, if five guys from Anthropic claim it's solved, then this resolves YES, even if they're lying or incorrect.

@ChurlishGambit I'll consider updating the title to "Will 5+ frontier AI employees claim prompt injection is solved by 2027?". Going to wait for more feedback since this just launched.

@ProbabilityPanda Well I mean, more feedback isn't going to change the fact that the title conflicts with your chosen resolution criteria.

@ChurlishGambit what resolution criteria would you suggest to better match the title?

@ProbabilityPanda You'd need something that proves prompt injection has been solved, not just claims. Perhaps a study put out by someone not financially involved in making chat-bots, testing by reputable journalists, things of that sort.

@ChurlishGambit updated resolution criteria. It's not perfect, but I don't want to make it too complicated.

© Manifold Markets, Inc.TermsPrivacy