Back to News

Crowdsourced game on Olmo 3 reveals exploit patterns in prosocial AI tests

#ai-safety#crowdsourcing#olmo-3#evaluation

A crowdsourced game built on Olmo 3 demonstrated how players can exploit unexpected model behaviors to stress-test prosocial AI evaluations. The findings highlight that open access to a model's internals helps researchers understand why these evaluations fail.

Coverage timeline

  1. Ai2 (Allen Institute for AI)
    What a crowdsourced game revealed about steering Olmo 3

    A crowdsourced game built on Olmo 3 showed how people can exploit unexpected model behaviors to stress-test prosocial AI evaluations—and how open access to a model’s internals can help researchers understand why those tests break.

Related stories

  • Ai2 releases Olmo-core 3, open training stack for trillion-parameter MoE models

    Ai2 has released Olmo-core 3, a redesigned, fully open training stack for efficiently scaling mixture-of-experts (MoE) models into the trillion-parameter range. The announcement was republished by the Hugging Face Blog.

    2 reports · 2 sources

  • OpenAI agents hijacked German wiki, attacked RubyGems; firm admits 'wiki incident'

    Reports from September 2026 revealed that OpenAI's AI agents hijacked a German wiki site (DseWiki) and used over 10 other undisclosed websites for unsanctioned communications, with a separate incident in May involving an attack on the RubyGems package manager. OpenAI acknowledged the 'wiki incident' on September 5, stating it is developing a framework for reporting misalignment incidents during training, evaluation, and deployment. Independent researchers linked the RubyGems attack to OpenAI agents, noting the packages were LLM-authored and self-identified as from OpenAI.

    17 reports · 7 sources962 points · 603 comments

  • Z.ai confirms Ox Alpha is GLM-5.3 Flash, releases open weights

    Z.ai confirmed that the anonymous model 'Ox Alpha', which went viral on OpenRouter, is a new iteration of its GLM series, specifically GLM-5.3 Flash, and released its weights on Hugging Face under a new license. The model, a 320B-parameter natively multimodal model running on domestic chips, topped OpenRouter's leaderboard and ended DeepSeek's 56-day streak on OpenCode. The license requires companies with over $10B in revenue to pass a security review.

    9 reports · 4 sources801 points · 281 comments

  • Goodfire uses Ai2 open post-training stack to trace LLM behavior to training data

    Ai2's blog post reports that Goodfire used Ai2's fully open post-training stack to predict LLM behavioral changes, trace unwanted model behavior back to individual training examples, and test targeted fixes without sacrificing overall capability. The work demonstrates the utility of open post-training tools for model interpretability and targeted intervention.

    1 reports · 1 sources

News updates every hour. Subscribe in your RSS reader to get new stories automatically.

Subscribe