About MisalignAI

MisalignAI was founded on a simple premise: as AI systems become more powerful and more deeply embedded in our lives, the consequences of their failures become more severe. Yet the AI industry lacks a centralized, data-driven record of these failures.

We aim to be the Bloomberg of AI safety - a trusted, independent source of record for AI incidents, model evaluations, and safety analysis. Our goal is not to slow down AI progress, but to ensure that progress is made responsibly, with full awareness of the risks and failures that occur along the way.

Every incident we document, every model we evaluate, and every analysis we publish is aimed at one outcome: safer AI systems that serve humanity's best interests.

Incidents in the database are verified against source material before publication. The severity scale is based on factors including direct harm, affected users, financial impact, and broader implications for AI safety.

Model safety scores combine public evidence signals, incident anchors, and a transparent methodology. Scores are directional research outputs, not guarantees of model behavior.

Dynamic APIs and contribution workflows are planned as separate services. The public frontend must continue to render even when those services are unavailable.

• 独立研究MisalignAI 的研究和分析独立于任何 AI 公司或投资方。我们不接受付费内容放置，不从被报道公司获取咨询费，不编辑内容由赞助商控制。
• 来源标准所有事件记录基于公开可验证的来源（新闻报道、法院文件、官方安全报告、学术研究）。每个事件都包含原始来源链接。
• 更正政策如果发现错误，我们承诺在 24 小时内更新内容并标注更正。更正历史在页面底部可见。
• 利益冲突披露如果有任何团队成员与报道内容存在利益冲突，将明确披露。
• 评分方法论模型评分基于公开可验证的基准测试结果和安全评估，不是付费评级。方法论详见相关文章链接。

• 事件数据库每周更新，重大事件 24 小时内添加
• 模型评分每月更新，新模型发布时即时评估
• 深度报告每季度发布
• 博客文章每周 2–3 篇

最后更新日期：2026-06-16