TL; DR 数据
10月17日，在一场旨在鼓励美国人在11月3日中期选举中提前投票的全国性竞选活动中，Meta中止了无国王联盟的官方Instagram账户。 美国各地正在发生近3000起事件。 东部时间下午1点半左右
Meta最初指责过度张贴，分享和评论。 次日，其通信副总裁安迪·斯通更改了解释。 Meta现在的声称，使用Claude AI的软件正在以4倍于正常速度读取Instagram配置文件，帖子和标签。
帐号在大约90分钟后再度被公之于众
Meta在全国投票活动中审查一个政治组织，指责它自己的自动化，然后声称它的自动化系统发现了另一个自动化工具。 Meta承认暂停是错误的，但其解释都没有从技术角度说明所发生的事。
Meta的回应正是我这样的独立技术专家所警告的。 因为我们不能让科技公司通过指责AI来逃避责任与责任。
技术公司除了自己都在责备一切 Meta是一个公司，技术可以推断一个7岁的孩子喜欢什么
这是我对Meta行动的独立技术分析，如果你支持或反对这个组织，你应该非常担心扎克伯格可以审查任何人，任何观点与Meta的商业利益或Zuckerberg的政治利益相冲突的人
Meta不停地改变它的解释
10月17日，安迪·斯通写道：
“当账户张贴、分享或评论频率极高时，可触发自动化系统。”
但组织者对此表示异议，他们说他们只刊登了少数文章，鉴于Meta自己的记录非常容易证明或反驳这一说法，我怀疑该公司需要更令人信服的解释。
一天后，斯通提供了不同的解释，说这一活动主要是该帐户的 " 阅读 " 行为。
张贴和阅读是完全不同的活动，一个胜任的开发者或数据科学家可以通过检查活动记录在几分钟内确定。 发表文章、分享内容和评论涉及向Instagram发送信息。 阅读简介、帖子和标签涉及要求Instagram提供信息。
Meta起初指责该组织的出版活动。 其工程师随后确定了阅读活性。
Stone应该解释什么证据支持他的第一次和第二次陈述。
Meta 责备Claude 不确认谁负责
Stone 称 Meta的安全系统在订正说明中检测到 " 使用Claude工具读取剖面图、帖子和标签 "。
Meta明确命名了Anthropic的AI软件，但没有识别该工具，由谁操作，或者是否由"无国王"组织者授权。
Meta是否建议禁止使用Claude来访问Instagram？如果是，则违反了哪些政策？如果没有，为什么Meta中止了该组织的帐户而不是限制自动化请求？
Meta 声称活性增加了四倍
Stone 称的请求率“在猛烈的暴发中急剧增加(~4x体积)。
四倍数？
如果某一账户通常在特定期间收到100个请求，则4倍即400个。 如果它通常收到10万人，则增加额意味着40万人。
我的读者人数一般在2000至25万，在社交媒体上，活动突然增加是正常的。
斯通提到 " 强烈暴发 "，这听起来令人震惊，但如果没有实际要求、时间框架和频率，则没有提供有意义的技术信息。 即使是少量的自动化请求也能以猛烈的暴发来。 Meta没有证实这些要求为何有理由暂停该组织的帐户。
读者计数和自动化请求是不同的测量，因此仅凭一个受众的大小无法确定Meta的安全门槛是否合理。
没有实际数字、测量期和Meta的安全阈值
无国王联盟正在协调近3 000起活动。 涉及其Instagram账户的活动增加是完全可预测的。
Meta使用一个百分比，但没有提供确定该活动是否有理由中止一个政治组织所需的数字。
Meta将自动读取描述为安全威胁
Stone 称的要求“反应了自动化的机器人交通，而不是普通的人类浏览我们的系统”。
我于2010年在硅谷创办了一家有风险支持的网络安全公司，并在2015至2019年间在Slack,HipChat,Messett,Telegram等平台上建造了世界上第一个安全聊天机——它们自动化了动作。 我有信心地说，自动化活动并非本质上的安全侵犯。 任何胜任的网络安全专业或聊天机器人开发者都不会自动将自动化归类为恶意。 这没有任何技术意义。 搜索引擎，监测工具，研究应用和AI助手每天每小时都会提出数十亿个自动化请求。 平台的系统旨在识别过多的流量，限制访问资格，并阻止违反其政策的请求。
Meta是否对使用自己的AI助手Muse实现与Facebook和Instagram互动自动化的人实施同样的限制？. 如果不是，为什么？
Meta承认停刊是错误的，但继续限制他们的账户
Stone声称，自动审查确认中止是错误的，大约一小时后恢复了账户。 这是不诚实的。 Meta恢复了公众的能见度，但组织者仍然被锁了好几个小时。 斯通说账户已经恢复，隐瞒了Meta仍然阻止该组织出版的事实。 即使在解释这种规模的错误时，Meta也选择了模糊了实际发生情况的语言。
无国王组织者报告说，他们的Instagram页面在大约90分钟后再次被公之于众，但他们仍然无法使用。 管理账户的人一直被锁在门外，直到东部时间下午5点33分。
Meta本身的自动审查已经认定停刊是错误的，然而组织者仍然无法在全国竞选期间发表帖子或与支持者沟通。
Stone later 称恢复行政准入需要更多时间。 为什么？我认为，唯一可信的解释是，恢复有意义的准入需要拥有必要授权的Meta雇员。 这意味着必须有人批准Meta已经承认的中止是错的。 Meta应查明是谁控制了这一决定，谁授权恢复，以及为什么组织者被锁在门外。
恢复其组织者仍然无法使用的Instagram页面并不能恢复他们的通信能力。 Meta决定对其进行审查
Meta的道歉并不能解释是谁做决定的。
Stone将最初的强制执行行动描述为“明显错误地采取”。 并不清楚。 我很清楚
自动系统无法独立决定哪些行为违反Instagram的政策
无国王联盟(No Kings Conference)是美国最突出的反特朗普运动之一，组织全国范围的示威活动，涉及上千个地点的上百万人。 在一次重大的动员运动中暂停其Instagram帐户是一项具有政治意义的决定，其后果是可预测的。
鉴于该组织的突出地位、时机和公众的反弹，多个Meta小组和高级管理人员必须参与有关中止和恢复的决定。 Zuckerberg本人本来会参与的，因为他对重大内容节制决定采取了亲身行动。
这不仅仅是软件错误。 Meta的自动系统执行这些决定。 恢复公众的能见度，同时将组织者锁在门外，这需要单独的执行决定，Meta还没有解释。
Meta的记录可以确定哪些人审查了事件，谁授权了修改，是否向高级主管作了简报，为什么恢复全面访问需要几个小时。
这不是最近第一个涉及Meta在选举期间审查政治通讯的案件。
Meta如何阻止现任总统(Luiz Inácio Lula da Silva)进行正常付费的Instagram升级，而他的算法和WhatsApp推荐他的对手Flávio Bolsonaro。 它的AI拒绝了Lula竞选活动的基本材料，但为Bolsonaro制作了相应的材料。 Meta的一些员工实时标出不平等待遇并升级；该公司后来提出与日期不符的转移技术借口
在选举期间，Meta跑出超过1600个政治广告，研究者发现这些广告是非法的，几乎都是用美元支付的，在选举年度，关于暂停账户的投诉增加了585%。 巴西案并未证实"无国王"停赛是协调或有意的。 它树立了Meta归因于技术故障的政治账户限制的又一例子。
Meta控制当政治组织和政党可以发表，广告，出现在建议中并与其受众沟通时。 对于引人注目的账户，Meta的主管负责做出这些决定，而不是自动系统。
Meta必须解释它的系统实际做了什么
Meta有记录绝对一切的日志。 他们确切了解导致暂停的原因、是谁或是什么导致的、施加了哪些限制以及哪些雇员对每一项行动负有责任。 也知道为何组织者在暂停赛后仍被禁赛。
Meta可能没有记录其雇员所作的每一项决定，但其系统日志载有解释发生情况所需的技术证据，除非这些记录被删除。
公众应该知道为什么马克·扎克伯格的公司正在审查政治组织和个人，其方式恰好有利于白宫的政治议程，在他们最需要沟通的时候就切断了他们的受众。
国会应当召集负责这些系统的工程师和数据科学家，让他们宣誓就职，并要求以系统日志为后盾的具体技术答案。 他们不应该被允许躲在公司声明背后
21个问题 Meta应该回答
Meta的电源
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
TL;DR
On 17 October, Meta suspended the official Instagram account of the No Kings Coalition during a nationwide campaign that was aimed at encouraging Americans to vote early in the 3 November midterm elections. Almost 3,000 events were taking place across the United States. Organisers lost access to their account at approximately 1:30pm Eastern Time, when they needed it to communicate with participants.
Meta initially blamed excessive posting, sharing, and commenting. The following day, its Vice President of Communications, Andy Stone, changed the explanation. Meta now claims that software using Claude AI was reading Instagram profiles, posts and hashtags at 4 times the normal rate.
The account became publicly visible again after approximately 90 minutes, but organisers couldn’t publish anything until Meta restored access hours later at 5:33pm.
Meta censored a political organisation during a national voting campaign, blamed its own automation, then claimed its automated systems had detected another automated tool. Meta admitted the suspension was wrong, but none of its explanations account for what happened from a technical standpoint.
Meta’s response is exactly what independent technical experts like me have warned about. I point this out because we can’t allow tech companies to avoid responsibility and accountability for their actions by blaming AI.
Tech companies are blaming everything but themselves. Meta is a company whose technology can infer what flavour of ice cream a 7-year-old likes from something her uncle’s dog posted on Instagram in 2020, yet apparently can’t explain why its own systems censored a political organisation at one of the most critical moments in its history.
This is my independent technical analysis of Meta’s actions, not an endorsement of the No Kings Coalition or its politics. It doesn’t matter if you support or oppose this organisation, you should be deeply concerned that Zuckerberg can censor anyone, including people or organisations you support, without providing evidence or accountability. The same filters can be used against anyone whose views conflict with Meta’s commercial or Zuckerberg’s political interests, now or in the future.
Meta keeps changing its explanation
On 17 October, Andy Stone wrote:
“When accounts post, share or comment at very high frequencies it can trigger automated systems.”
Stone blamed unusually frequent activity by the account itself, but the organisers disputed this, saying they’d published only a handful of posts and were livestreaming the events. Given how easily Meta’s own records could prove or disprove that claim, I suspect the company needed a more convincing explanation.
A day later, Stone provided a different explanation, saying the activity was primarily the account’s “reading” behaviour.
Posting and reading are completely different activities, something a competent developer or data scientist could identify within minutes by examining the activity logs. Publishing posts, sharing content and commenting involve sending information to Instagram. Reading profiles, posts and hashtags involves requesting information from Instagram.
Meta initially blamed the organisation’s publishing activity. Its engineers then identified reading activity instead.
Stone should explain what evidence supported his first and second statement.
Meta blamed Claude without identifying who was responsible
In his revised statement, Stone said Meta’s security systems had detected “the use of a Claude tool reading profiles, posts, and hashtags”.
Meta explicitly named Anthropic’s AI software but didn’t identify the tool, who operated it or whether the No Kings organisers had authorised it.
Is Meta suggesting that using Claude to access Instagram is prohibited? If so, which policy was violated? If not, why did Meta suspend the organisation’s account instead of restricting the automated requests?
Meta claims activity increased fourfold
Stone said the rate of requests “sharply increased (~4x volume) in intense bursts”.
Four times what?
If an account normally receives 100 requests during a particular period, four times that means 400. If it normally receives 100,000, the increase means 400,000.
My LinkedIn posts typically reach between 2,000 and 250,000 readers, while my posts on X range from 1,500 to 1.2 million. Sudden increases in activity are normal on social media.
Stone’s reference to “intense bursts” sounds alarming but provides no meaningful technical information without the actual request counts, timeframe and frequency. Even a small number of automated requests can arrive in intense bursts. What Meta hasn’t established is why those requests justified suspending the organisation’s account.
Reader counts and automated requests are different measurements, so the size of an audience alone cannot establish whether Meta’s security threshold was reasonable.
Without the actual numbers, the measurement period and Meta’s security thresholds, the figure doesn’t tell us anything.
The No Kings Coalition was coordinating almost 3,000 events. Increased activity involving its Instagram account was entirely predictable.
Meta is using a percentage without providing the numbers needed to determine if the activity justified suspending a political organisation.
Meta described automated reading as a security threat
Stone said the requests “resembled automated bot traffic rather than an ordinary human browsing our systems”.
I founded a venture-backed cybersecurity company in Silicon Valley in 2010 and built the first security chatbots in the world - they automated actions on platforms like Slack, HipChat, Messenger, and Telegram between 2015 and 2019. I can say with confidence that automated activity isn’t inherently a security violation. No competent cybersecurity professional or chatbot developer would automatically classify automation as malicious. That makes no technical sense. Search engines, monitoring tools, research applications and AI assistants make billions of automated requests every hour of every day. Platforms have systems that are designed to identify excessive traffic, restrict access credentials and block requests that violate their policies.
Does Meta apply the same restrictions to people using its own AI assistant, Muse, to automate interactions with Facebook and Instagram? If not, why?
Meta admitted the suspension was wrong but continued to restrict their account
Stone claimed an automated review recognised the suspension was wrong and restored the account approximately an hour later. This is disingenuous. Meta restored public visibility, but the organisers remained locked out for hours. By saying the account had been restored, Stone concealed the fact that Meta was still preventing the organisation from publishing. Even when explaining a mistake of this magnitude, Meta chose language that obscured what actually happened.
The No Kings organisers reported that their Instagram page became publicly visible again after approximately 90 minutes, but they still couldn’t use it. People managing the account remained locked out until 5:33pm Eastern Time.
Meta’s own automated review had already decided that the suspension was wrong, yet the organisers remained unable to publish posts or communicate with supporters during their nationwide campaign.
Stone later said that restoring administrative access took additional time. Why? The only credible explanation in my view, is that restoring meaningful access required a Meta employee with the necessary authority. That suggests someone had to approve the reversal of a suspension Meta had already admitted was wrong. Meta should identify who controlled that decision, who authorised the restoration and why the organisers were kept locked out.
Restoring an Instagram page that its organisers still can’t use doesn’t restore their ability to communicate. Meta’s decision to censor them continued affecting the organisation after its own systems had recognised the suspension was wrong.
Meta’s apology doesn’t explain who made the decisions
Stone described the original enforcement action as “clearly taken in error”. There’s nothing clear about it. It’s clear to me that the opposite to what he says is more likely to be true.
An automated system doesn’t independently decide which behaviour violates Instagram’s policies, especially when it comes to such prominent accounts.
The No Kings Coalition is one of America’s most prominent anti-Trump movements, organising nationwide demonstrations that involve millions of people across thousands of locations. Suspending its Instagram account during a major mobilisation campaign was a politically significant decision with predictable consequences.
Given the organisation’s prominence, the timing and the public backlash, multiple Meta teams and senior managers had to be involved in decisions about the suspension and restoration. Zuckerberg himself would have been involved, given his hands-on approach to major content moderation decisions.
This wasn’t simply a software error. People made the decisions, and Meta’s automated systems executed them. Restoring public visibility while keeping organisers locked out required separate enforcement decisions that Meta has yet to explain.
Meta’s records can identify which individuals reviewed the incident, who authorised changes, whether senior executives were briefed and why restoring full access took hours.
This isn’t the first recent case involving Meta censoring political communications during an election
I wrote article 18 hours ago to explain how in Brazil, Meta blocked the sitting president (Luiz Inácio Lula da Silva) from running normal paid Instagram promotions while its algorithms and WhatsApp recommended his opponent Flávio Bolsonaro. Its AI refused basic Lula campaign material but produced the equivalent for Bolsonaro. Some of Meta’s staff flagged the unequal treatment in real time and escalated it; the company later offered shifting technical excuses that don’t match the dates or the fact that Bolsonaro’s campaign kept using the same tools Lula couldn’t access.
During the election, Meta ran over 1,600 political ads that researchers found were illegal, almost all paid for in US dollars, and only reclassified or removed them after they’d already been shown up to 171.1 million times. Complaints about account suspensions increased 585% during the election year.The Brazil case doesn’t establish that the No Kings suspension was coordinated or deliberate. It establishes another example of politically consequential account restrictions that Meta attributed to technical failures.
Meta controls when political organisations and parties can publish, advertise, appear in recommendations and communicate with their audiences. For high-profile accounts, Meta’s executives are responsible for making those decisions, not automated systems.
Meta must explain what its systems actually did
Meta has logs that record absolutely everything. They know precisely what triggered the suspension, who or what caused it, which restrictions were imposed and which employees were responsible for each and every action. It also knows why organisers remained locked out after the suspension was reversed.
Meta might not document every decision made by its employees, but its system logs contain the technical evidence needed to explain what happened, unless those records have been deleted.
The public deserves to know why Mark Zuckerberg’s company is censoring political organisations and individuals in ways that just happen to favour the White House’s political agenda, cutting them off from their audiences precisely when they need to communicate most.
Congress should summon the engineers and data scientists responsible for these systems, put them under oath and demand specific technical answers backed by system logs. They shouldn’t be allowed to hide behind corporate statements or blame automation for decisions made by people.
21 questions Meta should answer
Meta’s eng
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.