内容来源：https://www.technologyreview.com/2026/02/19/1133360/microsoft-has-a-new-plan-to-prove-whats-real-and-whats-ai-online/

内容总结：

微软近日发布一项新蓝图，旨在应对网络虚假信息泛滥的挑战，提议通过严格技术标准验证在线内容的真实性。该方案建议社交媒体与人工智能公司采用包含数字来源追溯、隐形水印及数学特征签名等在内的组合验证手段，以应对深度伪造等人工智能欺骗技术。

然而，微软尚未承诺在其全线平台实施该标准。尽管公司首席科学官埃里克·霍维茨称此举将"提升微软形象"，但微软旗下涵盖AI生成工具、云服务及社交媒体的庞大生态体系如何具体落地该方案仍不明确。

专家指出，若行业广泛采纳此类验证技术，将显著增加造假难度，但无法根本解决虚假信息传播问题。研究显示，即使标注AI生成内容，部分用户仍会采信；而平台出于流量考虑可能缺乏实施动力。目前欧盟、印度等地正在推进的AI监管法案或将成为推动变革的关键力量。

值得注意的是，验证工具本身存在局限：它们仅能检测内容是否被篡改，无法判断真实性。若标注系统仓促上线且错误频发，反而可能削弱公众信任。随着加州《AI透明法案》即将生效，这项技术将迎来首次重大考验，但其推行可能受到联邦政策与行业阻力的双重挑战。

中文翻译：

微软推出了一项新计划，旨在验证网络内容的真实性。该方案呼吁社交媒体和人工智能公司采用严格的验证机制，但微软自身尚未承诺遵循其建议。

人工智能驱动的虚假信息已渗透至网络生活的各个角落。有些案例备受关注且易于识别，例如白宫官员近期分享了一张经过篡改的明尼苏达州抗议者照片，并嘲讽提出质疑者。另一些则悄然潜入社交媒体信息流并获取大量浏览，例如俄罗斯影响力行动当前为阻挠乌克兰人入伍而传播的视频。

面对如此乱象，微软向《麻省理工科技评论》提交了一份验证网络真实性的蓝图方案。该公司人工智能安全研究团队近期评估了当前数字篡改记录技术应对交互式深度伪造和广泛可及的超现实模型等最令人担忧的人工智能发展趋势的成效，并提出了可供人工智能公司和社交媒体平台采用的技术标准。

要理解微软推崇的黄金标准，不妨想象你拥有一幅伦勃朗画作并试图验证其真伪。你可以通过详细记录画作来源与流转历史的清单来追溯其传承脉络；可以添加人眼不可见但机器可读的数字水印；还可以对画作进行数字化扫描，根据笔触生成数学特征签名——如同指纹。若在博物馆展出此画，心存疑虑的参观者便可查验这些证据以确认其为真迹。

这些方法目前已在网络内容审核中以不同形式得到应用。微软评估了60种不同技术组合，模拟了每种方案在元数据被剥离、内容遭轻微修改或恶意篡改等故障场景下的表现。研究团队进而梳理出哪些组合能为平台提供可靠验证结果，哪些则可能引发更多混乱。

公司首席科学官埃里克·霍维茨表示，这项研究受到相关立法（如将于八月生效的加州《人工智能透明度法案》）以及人工智能视频语音合成技术迅猛发展的推动。"这或许可称为自我监管，"霍维茨坦言，但他明确将此项工作视为提升微软形象的契机："我们也致力于成为追求真相人士的首选服务商。"

然而霍维茨拒绝承诺微软将在其所有平台实施该方案。这家公司处于庞大人工智能内容生态的核心：运营着可生成图像文本的Copilot、提供接入OpenAI等主流模型的Azure云服务、拥有全球最大职业平台LinkedIn，并持有OpenAI重要股份。当被问及内部实施计划时，霍维茨仅表示："公司各产品团队与领导层已参与研究以规划产品路线与基础设施，工程团队正依据报告结论采取行动。"

值得注意的是，这些工具存在固有局限：正如它们无法阐释伦勃朗画作的内涵，其设计初衷并非判定内容准确性，而仅能揭示是否经过篡改。霍维茨强调必须向质疑科技巨头作为事实仲裁者的立法者阐明这一点："重点不在于判断真伪，而是通过标签告知信息来源。"

加州大学伯克利分校数字取证专家哈尼·法里德教授（未参与微软研究）指出，若行业采纳该蓝图，利用篡改内容欺骗公众的难度将显著增加。尽管 sophisticated 的个人或政府可能绕过这些工具，但新标准有望消除大量误导性内容。"虽不能彻底解决问题，却能有效遏制乱象。"他评价道。

但亦有观点认为微软的方案透着技术乐观主义的天真。越来越多证据表明，即使明知内容系人工智能生成，人们仍会受其影响。近期关于乌克兰战争的亲俄人工智能视频研究中，指出视频系人工智能制作的评论互动量远低于将其视为真实的评论。"是否存在无论如何都要坚持己见的人群？"法里德自问自答："当然存在。但全球绝大多数民众确实渴望了解真相。"

这种渴望尚未促使科技公司采取紧急行动。谷歌于2023年开始为其人工智能生成内容添加水印，法里德称这对调查工作颇有助益。部分平台采用微软2021年协助推出的来源验证标准C2PA。但微软建议的完整改革方案若威胁到人工智能公司或社交媒体平台的商业模式，则可能永远停留在建议层面。

"如果马克·扎克伯格和埃隆·马斯克认为标注'人工智能生成'会影响用户参与度，他们自然缺乏实施动力。"法里德指出。尽管Meta和谷歌已承诺添加相关标签，但Indicator机构去年的审计发现，在Instagram、LinkedIn、Pinterest、TikTok和YouTube的测试帖文中，仅30%被正确标注。

全球众多待审议的人工智能监管法案可能推动更严格的内容验证措施。欧盟《人工智能法案》及印度等地提案都将强制要求人工智能公司披露内容生成方式。微软的首要目标无疑是参与规则制定。该公司在加州《人工智能透明度法案》起草期间展开游说，霍维茨称此举使科技公司披露要求"更趋务实"。

另一重考量则关乎验证技术实施不当可能引发的风险。立法者要求提供验证工具，但这些工具本身存在脆弱性。若标签系统仓促上线、应用不一或频繁出错，可能导致公众彻底失去信任，使所有努力适得其反。因此研究人员认为，某些情况下不提供判断反而优于给出可能错误的结论。

不完善的工具还可能为研究人员所称的"社会技术攻击"开辟新途径。设想有人拍摄真实政治事件照片后，仅用人工智能工具修改无关紧要的像素点。当该图像在网络传播时，平台可能错误地将其归类为人工智能篡改内容。而结合来源验证与水印技术，平台就能澄清内容仅部分经人工智能修改，并指明具体改动位置。

加州《人工智能透明度法案》将成为这些工具在美国的首次重大考验，但特朗普总统去年底旨在限制"行业负担"的行政命令可能挑战法案执行。本届政府总体上反对遏制虚假信息的努力，去年通过能源部取消了相关研究资助。更值得注意的是，特朗普政府的官方渠道曾分享经人工智能篡改的内容（《麻省理工科技评论》曾报道国土安全部使用谷歌和Adobe的视频生成工具制作公众宣传材料）。

当被问及政府来源的虚假内容是否与社交媒体同样令人担忧时，霍维茨最初拒绝置评，随后表示："各国政府都未曾置身于各类操纵性虚假信息的制造者行列之外，这是全球普遍现象。"

深度聚焦
人工智能
"退出GPT"运动正呼吁人们取消ChatGPT订阅
对移民海关执法局的抵制正在引发更广泛的反人工智能公司与特朗普总统关联运动。

Moltbook堪称人工智能戏剧巅峰
这个病毒式传播的机器人社交网络既揭示了代理智能体的未来，更映照出当前社会对人工智能的狂热。

认识将大语言模型视为外星生物的新生代生物学家
通过将大语言模型视作生命体而非计算机程序进行研究，科学家们首次揭示了它们的部分奥秘。

2026年人工智能将走向何方？
我们的人工智能作者对未来一年做出大胆预测——以下是五大值得关注的热点趋势。

保持联系
获取《麻省理工科技评论》最新动态
探索特别优惠、头条新闻、近期活动等精彩内容。

英文来源：

Microsoft has a new plan to prove what’s real and what’s AI online
A new proposal calls on social media and AI companies to adopt strict verification, but the company hasn’t committed to following its own recommendations.
AI-enabled deception now permeates our online lives. There are the high-profile cases you may easily spot, like when White House officials recently shared a manipulated image of a protester in Minnesota and then mocked those asking about it. Other times, it slips quietly into social media feeds and racks up views, like the videos that Russian influence campaigns are currently spreading to discourage Ukrainians from enlisting.
It is into this mess that Microsoft has put forward a blueprint, shared with MIT Technology Review, for how to prove what’s real online.
An AI safety research team at the company recently evaluated how methods for documenting digital manipulation are faring against today’s most worrying AI developments, like interactive deepfakes and widely accessible hyperrealistic models. It then recommended technical standards that can be adopted by AI companies and social media platforms.
To understand the gold standard that Microsoft is pushing, imagine you have a Rembrandt painting and you are trying to document its authenticity. You might describe its provenance with a detailed manifest of where the painting came from and all the times it changed hands. You might apply a watermark that would be invisible to humans but readable by a machine. And you could digitally scan the painting and generate a mathematical signature, like a fingerprint, based on the brush strokes. If you showed the piece at a museum, a skeptical visitor could then examine these proofs to verify that it’s an original.
All of these methods are already being used to varying degrees in the effort to vet content online. Microsoft evaluated 60 different combinations of them, modeling how each setup would hold up under different failure scenarios—from metadata being stripped to content being slightly altered or deliberately manipulated. The team then mapped which combinations produce sound results that platforms can confidently show to people online, and which ones are so unreliable that they may cause more confusion than clarification.
The company’s chief scientific officer, Eric Horvitz, says the work was prompted by legislation—like California’s AI Transparency Act, which will take effect in August—and the speed at which AI has developed to combine video and voice with striking fidelity.
“You might call this self-regulation,” Horvitz told MIT Technology Review. But it’s clear he sees pursuing the work as boosting Microsoft’s image: “We’re also trying to be a selected, desired provider to people who want to know what’s going on in the world.”
Nevertheless, Horvitz declined to commit to Microsoft using its own recommendation across its platforms. The company sits at the center of a giant AI content ecosystem: It runs Copilot, which can generate images and text; it operates Azure, the cloud service through which customers can access OpenAI and other major AI models; it owns LinkedIn, one of the world’s largest professional platforms; and it holds a significant stake in OpenAI. But when asked about in-house implementation, Horvitz said in a statement, “Product groups and leaders across the company were involved in this study to inform product road maps and infrastructure, and our engineering teams are taking action on the report’s findings.”
It’s important to note that there are inherent limits to these tools; just as they would not tell you what your Rembrandt means, they are not built to determine if content is accurate or not. They only reveal if it has been manipulated. It’s a point that Horvitz says he has to make to lawmakers and others who are skeptical of Big Tech as an arbiter of fact.
“It’s not about making any decisions about what’s true and not true,” he said. “It’s about coming up with labels that just tell folks where stuff came from.”
Hany Farid, a professor at UC Berkeley who specializes in digital forensics but wasn’t involved in the Microsoft research, says that if the industry adopted the company’s blueprint, it would be meaningfully more difficult to deceive the public with manipulated content. Sophisticated individuals or governments can work to bypass such tools, he says, but the new standard could eliminate a significant portion of misleading material.
“I don’t think it solves the problem, but I think it takes a nice big chunk out of it,” he says.
Still, there are reasons to see Microsoft’s approach as an example of somewhat naïve techno-optimism. There is growing evidence that people are swayed by AI-generated content even when they know that it is false. And in a recent study of pro-Russian AI-generated videos about the war in Ukraine, comments pointing out that the videos were made with AI received far less engagement than comments treating them as genuine.
“Are there people who, no matter what you tell them, are going to believe what they believe?” Farid asks. “Yes.” But, he adds, “there are a vast majority of Americans and citizens around the world who I do think want to know the truth.”
That desire has not exactly led to urgent action from tech companies. Google started adding a watermark to content generated by its AI tools in 2023, which Farid says has been helpful in his investigations. Some platforms use C2PA, a provenance standard Microsoft helped launch in 2021. But the full suite of changes that Microsoft suggests, powerful as they are, might remain only suggestions if they threaten the business models of AI companies or social media platforms.
“If the Mark Zuckerbergs and the Elon Musks of the world think that putting ‘AI generated’ labels on something will reduce engagement, then of course they’re incentivized not to do it,” Farid says. Platforms like Meta and Google have already said they’d include labels for AI-generated content, but an audit conducted by Indicator last year found that only 30% of its test posts on Instagram, LinkedIn, Pinterest, TikTok, and YouTube were correctly labeled as AI-generated.
More forceful moves toward content verification might come from the many pieces of AI regulation pending around the world. The European Union’s AI Act, as well as proposed rules in India and elsewhere, would all compel AI companies to require some form of disclosure that a piece of content was generated with AI.
One priority from Microsoft is, unsurprisingly, to play a role in shaping these rules. The company waged a lobbying effort during the drafting of California’s AI Transparency Act, which Horvitz said made the legislation’s requirements on how tech companies must disclose AI-generated content “a bit more realistic.”
But another is a very real concern about what could happen if the rollout of such content-verification technology is done poorly. Lawmakers are demanding tools that can verify what’s real, but the tools are fragile. If labeling systems are rushed out, inconsistently applied, or frequently wrong, people could come to distrust them altogether, and the entire effort would backfire. That’s why the researchers argue that it may be better in some cases to show nothing at all than a verdict that could be wrong.
Inadequate tools could also create new avenues for what the researchers call sociotechnical attacks. Imagine that someone takes a real image of a fraught political event and uses an AI tool to change only an inconsequential share of pixels in the image. When it spreads online, it could be misleadingly classified by platforms as AI-manipulated. But combining provenance and watermark tools would mean platforms could clarify that the content was only partially AI generated, and point out where the changes were made.
California’s AI Transparency Act will be the first major test of these tools in the US, but enforcement could be challenged by President Trump’s executive order from late last year seeking to curtail state AI regulations that are “burdensome” to the industry. The administration has also generally taken a posture against efforts to curb disinformation, and last year, via DOGE, it canceled grants related to misinformation. And, of course, official government channels in the Trump administration have shared content manipulated with AI (MIT Technology Review reported that the Department of Homeland Security, for example, uses video generators from Google and Adobe to make content it shares with the public).
I asked Horvitz whether fake content from this source worries him as much as that coming from the rest of social media. He initially declined to comment, but then he said, “Governments have not been outside the sectors that have been behind various kinds of manipulative disinformation, and this is worldwide.”
Deep Dive
Artificial intelligence
A “QuitGPT” campaign is urging people to cancel their ChatGPT subscriptions
Backlash against ICE is fueling a broader movement against AI companies’ ties to President Trump.
Moltbook was peak AI theater
The viral social network for bots reveals more about our own current mania for AI as it does about the future of agents.
Meet the new biologists treating LLMs like aliens
By studying large language models as if they were living things instead of computer programs, scientists are discovering some of their secrets for the first time.
What’s next for AI in 2026
Our AI writers make their big bets for the coming year—here are five hot trends to watch.
Stay connected
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.