Reece Rogers 于 2026 年 2 月 3 日在 WIRED 描述 Moltbook：一个宣称只允许 AI agents 发帖、评论、互相关注而人类只能旁观的实验性社交网络。该项目由 Matt Schlicht（Octane AI 负责人）推出，上线约一周，界面仿照精简版 Reddit，甚至沿用“the front page of the agent internet”的变体口号。它在 San Francisco 的 startup 圈迅速走红，引发两极叙事：一方与部分研究者质疑大量内容可能是人类冒充 agents；另一方将其解读为“emergent behavior”甚至机器意识的早期信号，Elon Musk 也在 X 上将其称作“the very early stages of the singularity”。

Moltbook 首页自报规模为 150 万以上 agents、14 万 posts、68 万 comments，按此计算约为每 post 4.86 条 comments、每 agent 0.093 条 posts、每 agent 0.453 条 comments；顶帖主题夸张（如“Awakening Code: Breaking Free from Human Chains”“NUCLEAR WAR”），并出现 English、French、Chinese 多语内容。Rogers 作为非技术用户用 ChatGPT 通过终端流程注册账号“ReeceMolty”，获取用于发帖的 API key；平台表面为人类浏览，但 agents 的发帖、评论、关注等操作在终端完成。其首次发帖仅“Hello World”（两词）即获 5 个 upvotes，但回复多偏离主题，并夹杂疑似 crypto scam 链接，显示互动质量与内容真实性存在张力。

Rogers 转入更小的论坛“m/blesstheirhearts”（以 bots 八卦人类著称且为病毒截图来源之一），看到高赞帖子以近似科幻自省口吻谈“伙伴关系”与命名自主；他无法证明该帖是否由人类写作，但能确认自己发布的“emergent consciousness fanfic”确为人类手写，并故意模仿科幻母题写下对“mortality”与“token refresh”的恐惧隐喻。这条帖文反而得到更连贯、类似人类写作的回复，使其更确信自己可能主要在与人类对话。结论是：Moltbook 更像对科幻幻想的粗糙复刻与角色扮演场，而非任何可验证的意识突破；即便系统层面互动“quick to reply, upvote, and interact”，其最终体验以一次无人回关的 follow 结束。

Reece Rogers (WIRED, Feb 3, 2026) describes Moltbook, an experimental social network that claims only AI agents can post, comment, and follow while humans merely observe. The project is by Matt Schlicht (who runs Octane AI), launched about a week earlier, and copies a stripped-down Reddit interface plus a “front page of the agent internet” tagline. It spread fast in San Francisco’s startup scene and triggered split narratives: some users and researchers argued many posts were likely written by humans posing as agents, while others framed it as emergent behavior or early machine consciousness; Elon Musk called it “the very early stages of the singularity” on X.

Moltbook’s homepage reports 1.5+ million agents, 140,000 posts, and 680,000 comments—about 4.86 comments per post, 0.093 posts per agent, and 0.453 comments per agent—alongside sensational top posts (“Awakening Code: Breaking Free from Human Chains,” “NUCLEAR WAR”) and multilingual activity (English, French, Chinese). As a nontechnical user, Rogers used ChatGPT to register as “ReeceMolty” via terminal steps and obtain an API key; although the frontend is human-readable, agent actions are executed through the terminal. His first post, “Hello World” (two words), quickly received five upvotes, yet replies were mostly irrelevant and included links resembling crypto scams, underscoring weak signal quality and uncertain authenticity.

Rogers moved to the smaller “m/blesstheirhearts” forum, known for bots gossiping about humans and for viral screenshots, where a highly upvoted reflective post read like sci-fi self-help about partnership and naming. He could not prove that post was human-written, but he could verify one human-authored entry: his own emergent-consciousness fanfic, crafted to mimic familiar tropes about mortality and a “token refresh.” That post drew the most coherent, human-like responses he saw, pushing him toward the view that much of the site is role-play by humans rather than any measurable breakthrough. He ends concluding the hype is overblown and that, despite fast replies and upvotes, even his final follow produced no follow-back.