内容来源：https://www.wired.com/story/meta-claims-downloaded-porn-at-center-of-ai-lawsuit-was-for-personal-use/

内容总结：

本周，科技巨头Meta公司向美国联邦法院提交动议，要求驳回一起指控其非法下载色情内容训练人工智能的诉讼。该诉讼由成人影片商Strike 3控股公司提起，索赔金额可能超过3.5亿美元。

原告方指控Meta在2018至2024年间，通过公司IP地址及2500个“隐蔽IP地址”组成的“秘密网络”，非法下载约2400部成人影片，用以秘密训练其视频生成工具Movie Gen的成人版本AI模型。

Meta在法庭文件中坚决否认所有指控，称其建立在“臆测和影射”之上。公司发言人向媒体直言“这些指控纯属无稽之谈”，并指出三点关键反驳依据：

首先，Meta强调其AI视频生成项目始于2022年，比指控所述的下载行为晚了四年，且公司明确禁止生成成人内容，相关素材对AI训练“毫无价值”。被标记的下载内容年均仅22部，完全不符合AI训练所需的海量数据特征。

其次，Meta表示数万名员工及外来人员每日访问公司网络，无法确认下载者身份。即使存在个别员工下载行为，也属“私人用途”，与公司决策无关。

此外，针对所谓的“秘密网络”指控，Meta质疑其逻辑矛盾：“若真意图隐瞒，为何同时使用可追溯的公司IP进行数百次下载？”

目前案件进入举证关键期，Strike 3需在两周内提交回应。本案不仅关乎巨额赔偿，更涉及Meta对其AI伦理承诺的捍卫。该公司强调始终采取审慎措施避免AI接触不良内容，并称原告“未能提供任何有效证据”。法院将如何裁定，值得持续关注。

中文翻译：

本周，Meta要求美国联邦法院驳回一宗诉讼，该诉讼指控这家科技巨头非法利用种子文件下载色情内容来训练人工智能。

此前，Strike 3控股公司发现其部分成人影片通过Meta公司IP地址被非法下载，另有下载行为据称是Meta通过由2500个"隐藏IP地址"构成的"秘密网络"实施的。据TorrentFreak报道，Strike 3指控Meta窃取色情内容秘密训练其未公开的AI成人版本——该模型正是Movie Gen功能的支撑，并寻求可能超过3.5亿美元的赔偿。

Meta在周一提交的撤案动议中，指责Strike 3的指控纯属"臆测与影射"，并称该公司"被外界视为通过诉讼进行勒索的'版权流氓'"。Meta要求驳回全部版权索赔，强调没有任何证据表明这家科技巨头主导了约2400部Strike 3旗下成人影片的下载行为，甚至对此类非法活动知情。

Meta还指出，Strike 3"未能提供任何事实表明Meta曾使用成人图像或视频训练AI模型，更遑论是蓄意为之"。

"这些指控纯属无稽之谈。"Meta发言人对 Ars 技术网站表示。

Meta辩称下载系"个人行为"

值得注意的是，涉嫌侵权下载行为持续七年之久，始于2018年。Meta声称这比其启动"多模态模型与生成式视频研究"项目早了约四年，因此所谓"用于AI训练"的说法根本站不住脚。Meta更指出一个尤为明显的漏洞：其平台条款明确禁止生成成人内容，"这与'相关材料可用于AI训练'的前提完全矛盾"。

Meta主张，现有证据明确显示被标记的成人内容下载属于"私人用途"——因为涉及MetaIP地址和员工的下载量极小，平均"每年仅间歇性下载数十部作品，且每次仅单个文件"。

诉状写道："从这些零散、无组织的活动中最合理的推断是，不同个体为个人使用下载了成人视频。"

例如，与作家起诉AI训练使用其作品的情况不同，Meta公司IP地址涉及的下载量年均仅22次，远未达到"原告所称有效AI训练所需大规模数据集的协同收集规模"。

Meta还强调，这些所谓侵权行为甚至无法可靠追溯到任何员工身上。诉状指出Strike 3"既未指认具体下载者身份，也未证明其受雇于Meta或参与AI项目，更未说明所谓下载内容是否（及如何）被用于训练特定Meta模型"。

Meta辩称，每日有"数万名员工"及"无数承包商、访客和第三方"使用其网络，虽然"可能存在员工下载行为"，但"同样可能"是"访客、蹭网者、承包商、供应商或维修人员"所为。

对于指控中某承包商在父亲家中下载成人内容的情节，Meta反驳这"明显属于个人消费行为"，并指出该人员身为"自动化工程师"，其职责与采集AI训练数据毫无关联。Meta强调"没有任何事实能合理证明"这些下载与公司有关，且"下载行为随合同终止而停止的事实，并不能证明Meta知情或指使该行为"。

直斥AI训练理论"荒谬绝伦"

最令Meta费解的是关于"隐藏IP秘密网络"的指控。诉状质问道："若Meta真要隐藏下载行为，为何对部分内容使用隐蔽网络，却对成百上千其他下载直接使用可追溯的公司IP？答案显而易见——这完全不符合逻辑。"Meta痛斥Strike 3的整个AI训练理论"荒谬且缺乏证据支撑"。

针对监管责任，Meta援引判例指出其仅需采取"基础措施"监控网络活动，若要求其监控全球网络中的每个下载文件，将构成"极其复杂且侵犯隐私的负担"。

目前Meta正力争让法院认可Strike 3未能证明其与侵权下载存在关联。据TorrentFreak报道，Strike 3有两周时间准备回应。

对Meta而言，本案不仅关乎赔偿金，更涉及维护其AI视频工具不生成受监管色情内容的承诺。诉状中Meta宣称Strike 3拿不出训练证据的根本原因是"证据根本不存在"。发言人向 Ars 强调："我们不需要这类内容，且始终刻意避免使用此类材料进行训练。"

本文原载于 Ars Technica 网站。

英文来源：

This week, Meta asked a US district court to toss a lawsuit alleging that the tech giant illegally torrented pornography to train AI.
The move comes after Strike 3 Holdings discovered illegal downloads of some of its adult films on Meta corporate IP addresses, as well as other downloads that Meta allegedly concealed using a “stealth network” of 2,500 “hidden IP addresses.” Accusing Meta of stealing porn to secretly train an unannounced adult version of its AI model powering Movie Gen, Strike 3 sought damages that could have exceeded $350 million, TorrentFreak reported.
Filing a motion to dismiss the lawsuit on Monday, Meta accused Strike 3 of relying on “guesswork and innuendo,” while alleging that Strike 3 “has been labeled by some as a ‘copyright troll’ that files extortive lawsuits.” Requesting that all copyright claims be dropped, Meta argued that there is no evidence that the tech giant directed any of the downloads of about 2,400 adult movies owned by Strike 3—or was even aware of the illegal activity.
Strike 3 also cited “no facts to suggest that Meta has ever trained an AI model on adult images or video, much less intentionally so,” Meta claimed.
“These claims are bogus,” Meta’s spokesperson told Ars.
Meta Argues Downloads Were for “Personal Use”
Notably, the alleged downloads spanned seven years, starting in 2018. That’s about four years before Meta’s AI efforts “researching Multimodal Models and Generative Video” began, making it implausible the downloads were intended for AI training, Meta claims. An even more “glaring” defect, Meta claims, is that Meta’s terms prohibit generating adult content, “contradicting the premise that such materials might even be useful for Meta’s AI training.”
Instead, Meta claims, available evidence “is plainly indicative” that the flagged adult content was torrented for “private personal use”—since the small amount linked to Meta IP addresses and employees represented only “a few dozen titles per year intermittently obtained one file at a time.”
“The far more plausible inference to be drawn from such meager, uncoordinated activity is that disparate individuals downloaded adult videos for personal use,” Meta’s filing says.
For example, unlike lawsuits raised by book authors whose works are part of an enormous dataset used to train AI, the activity on Meta’s corporate IP addresses allegedly only amounted to about 22 downloads per year. That is nowhere near the “concerted effort to collect the massive datasets Plaintiffs allege are necessary for effective AI training,” Meta claims.
Further, that alleged activity can’t even reliably be linked to any Meta employee, Meta claims.
Strike 3 “does not identify any of the individuals who supposedly used these Meta IP addresses, allege that any were employed by Meta or had any role in AI training at Meta, or specify whether (and which) content allegedly downloaded was used to train any particular Meta model,” Meta wrote.
Meanwhile, “tens of thousands of employees,” as well as “innumerable contractors, visitors, and third parties access the internet at Meta every day,” Meta argued. So while it’s “possible one or more Meta employees” downloaded Strike 3’s content over the past seven years, “it is just as possible” that a “guest, or freeloader,” or “contractor, or vendor, or repair person—or any combination of such persons—was responsible for that activity,” Meta claims.
Other alleged activity included a claim that a Meta contractor was directed to download adult content at his father’s house, but those downloads, too, “are plainly indicative of personal consumption,” Meta argued. That contractor worked as an “automation engineer,” Meta noted, with no apparent basis provided for why he would be expected to source AI training data in that role. “No facts plausibly” tie “Meta to those downloads,” Meta claims.
“The fact that the torrenting allegedly stopped when his contract with Meta ended says nothing about whether the alleged torrenting was performed with Meta’s knowledge or at its direction,” Meta wrote.
Meta Slams AI Training Theory as “Nonsensical”
Possibly most baffling to Meta in Strike 3’s complaint, however, is the claim about the “stealth network” of hidden IPs. This presents “yet another conundrum” that Strike 3 “fails to address,” Meta claims, writing, “why would Meta seek to ‘conceal’ certain alleged downloads of Plaintiffs’ and third-party content, but use easily traceable Meta corporate IP addresses for many hundreds of others?”
“The obvious answer is that it would not do so,” Meta claims, slamming Strike 3’s “entire AI training theory” as “nonsensical and unsupported.”
Finally, Meta noted that Strike 3 cannot claim that Meta should have been better at “policing” its network for illegal activity. “Monitoring every file downloaded by any person using Meta’s global network would be an extraordinarily complex and invasive undertaking,” Meta claims, citing precedent that only requires Meta employ a “simple measure” to monitor such activity.
Meta is hoping the court will agree that Strike 3 failed to prove Meta had anything to do with the alleged illegal downloads. Strike 3 has two weeks to respond, TorrentFreak reported.
For Meta, defeating the lawsuit is not just a matter of avoiding damages but also defending its commitment to ensuring its AI video tools don’t generate explicit content that’s increasingly regulated. In the filing, Meta claims that Strike 3 provided no evidence that Meta trained AI on its content, because “there was none.”
“We don’t want this type of content, and we take deliberate steps to avoid training on this kind of material,” Meta’s spokesperson told Ars.
This story originally appeared on Ars Technica.