Trump admin's retreat from foreign-influence fight is weakening US defenses, sources say

Erik S. Lesser / AFP via Getty Images

States and industry report getting less help from the federal government ahead of the midterms—even as AI gives foreign operatives a boost.

The Trump administration’s retreat from programs that track foreign influence operations has weakened various cross-sector relationships used to identify and confront those campaigns as midterm elections fast approach — even as artificial intelligence makes those operations easier to deploy — multiple sources told Nextgov/FCW .

State officials say they are receiving fewer election-specific federal briefings than in previous cycles, while others who track foreign influence have observed routine exchanges with government and technology companies fall away. Some also note that federal workforce departures have weakened cooperation with outside partners, including social media platforms, to track false information online.

While the federal government continues to gather intelligence on foreign threats to U.S. elections, some of those who lean on those findings say federal agencies are less reliably sharing information with non-federal researchers and state officials responsible for countering those campaigns. That cooperation began to weaken before President Donald Trump returned to office, and his administration has since dismantled several programs dedicated to countering the problem.

This story is based on interviews with current and former federal officials, state election officials and specialists in cybersecurity and influence operations. Several people spoke on the condition of anonymity to be candid about their views and for fear of retribution. Their concerns come as fast-evolving AI tools now allow foreign influence operators to do more than just generate fake images or audio, instead helping them produce plausible-sounding messages, find audiences to target and autonomously adjust campaigns at a scale not seen before.

The Cybersecurity and Infrastructure Security Agency referred questions about foreign malign influence to the Justice Department, saying that influence operations fall outside CISA’s mission. DOJ did not immediately respond to a request for comment.

“Election security is national security, and protecting our critical infrastructure remains a top priority for the Trump Administration, DHS, and CISA,” a CISA spokesperson said. “CISA works closely with intelligence and law enforcement partners to monitor threats to U.S. critical infrastructure, including election infrastructure.”

The spokesperson added that the agency is committed to supporting state and local election officials “to protect election infrastructure and safeguard our democracy.”

The Office of the Director of National Intelligence did not respond to a request for comment.

Foreign influence operations aim to shape what people believe and how they act, often without revealing who is behind the message. Around elections, that can mean spreading false voting information, stoking political divisions or casting doubt on results. Such messages can be harder to question when they appear to come from or be amplified by fellow Americans and reinforce views people already hold.

One election nonprofit official told Nextgov/FCW that, during the November 2024 election, they received an early-morning text from a colleague in Georgia asking about bomb threats at voting sites, which U.S. officials later deemed may have been tied to Russia . The official was sitting in CISA’s election operations room. Finding out whether other jurisdictions were facing similar threats meant simply walking across the room and asking federal personnel.

“Any time you have federal support to help connect the dots, it’s a game-changer,” the official said, noting that they have not been alerted to or invited to a CISA elections operations room for the coming midterms.

The CISA spokesperson said that, before and during Election Day, the agency will “activate an Incident Coordination Group to establish a unified operational picture, rapidly share critical information, coordinate essential services for stakeholders in the field and strengthen incident response within its mission and authorities.”

Bret Schafer, senior director of policy and research for information operations at the Institute for Strategic Dialogue, said efforts to bring government, technology companies and researchers together grew during Trump’s first term, after Russia’s interference attempts in the 2016 election exposed gaps in their understanding of foreign campaigns.

Each could see activity the others could not. Sharing those findings helped them recognize operations that might otherwise go unnoticed, Schafer said last week on a panel at a national-security law conference hosted by the American Bar Association.

Those routine exchanges with the government have become harder to sustain. Now, “civil society is out on both sides,” he said, adding that he had gone at least two years without an information-sharing conversation with a relevant technology company, and a couple of years without one involving the government.

Much of that cooling began in the Biden administration, when GOP litigation over federal interactions with social media companies made participants more cautious . The Trump administration has gone further. It shuttered CISA work focused on election disinformation, disbanded the FBI’s Foreign Influence Task Force and closed the State Department’s office for countering foreign information threats, all largely on censorship grounds. The ODNI Foreign Malign Influence Center also faced a similar fate last August under then-director Tulsi Gabbard.

An office can retain its responsibilities while losing the relationships that helped it do the work, according to a person familiar with U.S. efforts to address influence operations. But someone who once knew exactly whom to call may find that their counterpart has departed, or that the people still there are unsure whether leadership supports exchanging information.

One current U.S. official described an environment that puts less public emphasis on election security at CISA and across the Department of Homeland Security than in previous cycles, with greater attention to other critical infrastructure matters, including water systems .

The official speculated that employees could hesitate to highlight their election security work after seeing others who worked on elections lose their jobs, though they noted they had not heard staff express that fear directly. The official also acknowledged that midterms generally receive less attention than presidential elections.

The administration has maintained that federal support continues. ODNI has assigned officials to coordinate election threat intelligence, while CISA outlined election support plans last month. Defense Secretary Pete Hegseth also directed Cyber Command and military intelligence agencies to prioritize foreign election threats and coordinate with the Department of Homeland Security last month.

A current Defense Department analyst argued that closing offices focused on tracking influence operations does not necessarily erase expertise, because personnel can continue the work elsewhere in government or as contractors. But it’s unclear how much of that work has continued, or whether those who left government still have access to relevant information and federal contacts they once relied on.

The analyst recalled a strong working relationship with Meta, which operates Facebook and Instagram, when they worked on those issues around two years ago. The analyst could not say whether those exchanges continue today.

Similarly, one former U.S. official recalled preparations for the 2020 election beginning roughly a year before Americans went to vote, with coordination growing more frequent as Election Day approached. Some calls included 80 to 100 participants, including private companies, discussing topics that included ballot security and broadcasting accurate voting information.

“Our nation’s security is best protected when private individuals, private sector, industry and government work together toward a common goal to protect our freedom and way of life, and that includes reporting suspicious and illegal activity and leveraging legal access and authorities to protect the American people and interests,” Erin Joe, special counsel at Wiley Rein LLP and a former senior FBI executive, told Nextgov/FCW in an interview.

At the ABA conference, former DHS Secretary Michael Chertoff argued that the government still has a role in helping people recognize deception. Officials can explain where a message came from or show that a supposed recording was generated by AI, without telling people which political views to accept.

“It is fighting bad speech with good speech,” Chertoff said.

A TikTok spokesperson referred Nextgov/FCW to a webpage outlining steps the company is taking to prepare for midterms and disrupt covert influence operations on its platform. Several other social media companies and AI firms — including Meta, X, Anthropic and OpenAI — did not respond to requests for comment.

Two state election officials described how their offices are responding to misleading information while receiving fewer election-specific federal briefings than in previous cycles.

One said identifying who is behind false messages can be difficult. Their office often focuses on information that could confuse people about when, where and how to vote, as well as threats to election workers.

When a misleading claim surfaces, communications staff and leadership decide whether responding would help voters or draw more attention to something that might otherwise fade away. Threats to election officials are referred to law enforcement, the official said.

The same official said they had not received an election-specific federal briefing “in quite some time,” although they attend regularly held CISA calls about software vulnerabilities and known flaws being exploited by hackers. Those discussions offer little information that is specific to elections, the official said. Monthly meetings through the state’s fusion center still bring together state, local and federal law enforcement to discuss threats to election officials.

The second official said their office sends a representative to federal election-threat briefings whenever they are available, but those briefings have occurred less frequently than in 2024.

That official nevertheless described their election offices as better prepared than in past years, with multiple backup plans, in case disruptions occur.

“We hope that we don’t have to use them. But if we do, they’re there,” the official said.

Those preparations come as AI has lowered the cost and effort required to conduct influence campaigns. AI has played a key role in helping to run these operations, from writing posts and distributing them to studying how people respond, said Snehal Antani, CEO and co-founder of cybersecurity company Horizon3.ai. Operators can test different messages and use the results to adjust their approach, and AI agents can automate much of that work, reducing the time and expertise it once required, he said.

“AI has dramatically reduced the skills required and dramatically accelerated the speed at which this cycle can be executed,” Antani said.

Another former U.S. official added that increasingly capable, downloadable open-source models allow smaller actors to undertake work that once required substantial investment and specialized engineering.

“The barrier to entry is much lower,” the former official said, adding that AI-generated text can blend into everyday online conversations and make it harder to spot than a manipulated image or video with visible flaws.

Russia is more likely to pursue more obvious inflammatory messaging this cycle, while China’s efforts may be more covert to serve longer-term objectives, said another current official, adding that both are expected to make greater use of AI. Pro-Iran outfits have especially leveraged AI more overtly in recent months amid the war that began in late February.

But the official cautioned that more powerful AI tools and more convincing content do not necessarily mean a foreign operation is changing how people vote. An image or video getting lots of views does not mean people believe it or will vote differently because of it, they added.

In some cases, an influence campaign may not even target eligible voters. For instance, content aimed at younger Americans could sway their political stances, but they may not be of legal age to cast a ballot.

And in other settings, a foreign influence operation may not be needed for an American to vote a certain way.

“If I see gas is $7 a gallon, I don’t necessarily need a lot more to go vote blue,” the official said, referring to California voters who have faced record-high fuel prices in some parts of the state.

These foreign threats come as political campaigns at home have increasingly used AI in their advertisements.

The growing use of AI-generated political videos could also make it harder to distinguish real footage from fabricated material, according to research released Wednesday by cybersecurity firm Check Point.

“Video is one of the easier ways to make a claim feel real, since people tend to trust what they can see and hear for themselves,” the company’s findings say. As AI-generated political content becomes more common, it becomes harder for voters to “trust that a video of a candidate is real” and easier for anyone “to dismiss a real video as fake.”

The scale and sophistication of false information today “makes our sensitivity to it a little bit more dull than perhaps in years past,” said Kimberly Kearney, a former CIA case officer.

“We as individuals have to have the desire to know whether what we’re consuming is true or false, or if it feeds into a narrative that is convenient or more comfortable for us,” she said in an interview. “So it starts with that cognitive honesty and awareness to want to know whether what you’re consuming is accurate or if it could have a bias to it.”

NEXT STORY: US Army to stage counter-drug training exercises in South America

Erik S. Lesser / 法新社/盖蒂图片社

各州和行业报告称，在中期选举前夕，他们从联邦政府获得的帮助越来越少——尽管人工智能为外国特工提供了助力。

多位消息人士告诉 Nextgov/FCW，随着中期选举的临近，特朗普政府退出追踪外国影响力行动的项目，削弱了用于识别和对抗这些行动的各种跨部门关系——尽管人工智能使这些行动更容易部署。

州政府官员表示，他们收到的与选举相关的联邦简报比以往选举周期要少，而其他追踪外国影响的人士也注意到，政府与科技公司之间的例行交流正在减少。一些人还指出，联邦政府工作人员的离职削弱了与外部合作伙伴（包括社交媒体平台）在追踪网络虚假信息方面的合作。

尽管联邦政府仍在收集有关外国势力威胁美国选举的情报，但一些依赖这些情报的人士表示，联邦机构与负责应对此类威胁的非联邦研究人员和州政府官员之间的信息共享越来越不可靠。这种合作在唐纳德·特朗普总统连任之前就开始减弱，而他的政府此后又解散了多个旨在应对这一问题的项目。

本文基于对现任和前任联邦官员、州选举官员以及网络安全和影响力行动专家的采访。部分受访者因担心遭到报复而要求匿名，以便坦诚表达观点。他们的担忧源于快速发展的人工智能工具，这些工具如今不仅能让外国影响力行动者生成虚假图像或音频，还能帮助他们制作听起来合情合理的宣传信息，找到目标受众，并以前所未有的规模自主调整宣传活动。

网络安全和基础设施安全局（CISA）将有关外国恶意影响的问题转给了司法部，称此类影响行动不属于CISA的职责范围。司法部尚未对此置评。

“选举安全就是国家安全，保护我们的关键基础设施仍然是特朗普政府、国土安全部和网络安全与基础设施安全局（CISA）的首要任务，”CISA发言人表示。“CISA与情报和执法伙伴密切合作，监控美国关键基础设施（包括选举基础设施）面临的威胁。”

发言人补充说，该机构致力于支持州和地方选举官员，“以保护选举基础设施并维护我们的民主”。

国家情报总监办公室未回应置评请求。

外国影响力行动旨在塑造人们的信仰和行为方式，而且往往不会透露幕后操纵者的身份。在选举期间，这可能意味着散布虚假的投票信息、煽动政治分裂或质疑选举结果。当这些信息看似来自美国民众或被美国民众放大，并强化了人们已有的观点时，人们就更难质疑这些信息。

一位选举非营利组织的官员告诉Nextgov/FCW，在2024年11月大选期间，他清晨收到一位佐治亚州同事发来的短信，询问投票站的炸弹威胁情况。美国官员后来认为这些威胁可能与俄罗斯有关。当时这位官员正坐在美国网络安全和基础设施安全局（CISA）的选举行动室里。要了解其他地区是否面临类似的威胁，只需走到房间另一边询问联邦工作人员即可。

“任何时候，只要有联邦政府的支持来帮助理清头绪，就能改变游戏规则，”这位官员说，并指出他们尚未收到有关即将到来的中期选举的 CISA 选举操作室的通知或邀请。

美国网络安全和基础设施安全局 (CISA) 发言人表示，在选举日之前和选举日期间，该机构将“启动一个事件协调小组，以建立统一的行动态势图，快速共享关键信息，协调为现场利益相关者提供的基本服务，并在其使命和权限范围内加强事件响应”。

战略对话研究所信息作战政策与研究高级主管布雷特·沙弗表示，在俄罗斯干预 2016 年大选暴露了政府、科技公司和研究人员对外国竞选活动的理解存在差距之后，在特朗普的第一任期内，将他们聚集在一起的努力日益增多。

每个人都能看到其他人看不到的活动。谢弗上周在美国律师协会主办的国家安全法律会议的一个小组讨论会上说，分享这些发现帮助他们识别出一些原本可能被忽视的行动。

与政府的例行交流如今越来越难以维持。他表示，现在“公民社会已经两边都退出了”，并补充说，他至少两年没有与相关科技公司进行过信息共享对话，而与政府进行过此类对话也已有两年之久。

这种降温趋势很大程度上始于拜登政府时期，当时共和党就联邦政府与社交媒体公司的互动提起诉讼，使得参与者更加谨慎。特朗普政府则走得更远。它关闭了网络安全和基础设施安全局（CISA）负责打击选举虚假信息的工作，解散了联邦调查局（FBI）的外国影响工作组，并关闭了国务院负责应对外国信息威胁的办公室，所有这些举措主要都是以审查为由。去年8月，在时任局长图尔西·加巴德的领导下，国家情报总监办公室（ODNI）的外国恶意影响中心也面临类似的命运。

据一位熟悉美国应对影响力行动的人士透露，一个机构可能在失去赖以开展工作的关系的同时，仍然保留其职责。但曾经确切知道该联系谁的人可能会发现，他们的对接人已经离职，或者仍在职的人不确定领导层是否支持信息交流。

一位现任美国官员表示，与以往的选举周期相比，网络安全和基础设施安全局 (CISA) 以及整个国土安全部对选举安全的重视程度有所降低，而对其他关键基础设施问题（包括供水系统）的关注度则更高。

这位官员推测，员工们看到其他参与选举安保工作的同事因此丢掉工作，可能会不愿公开自己的工作，不过他也指出，自己并未直接听到员工表达过这种担忧。这位官员还承认，中期选举通常不如总统选举那样受到关注。

政府方面一直强调联邦政府将继续提供支持。国家情报总监办公室已指派官员协调选举威胁情报工作，而网络安全和基础设施安全局（CISA）上个月也概述了选举支持计划。国防部长皮特·赫格塞斯上个月也指示网络司令部和军事情报机构优先应对外国选举威胁，并与国土安全部协调应对。

一位现任国防部分析师认为，关闭专门负责追踪影响力行动的机构并不一定会抹杀专业知识，因为相关人员可以在政府其他部门或以承包商的身份继续从事这项工作。但目前尚不清楚有多少此类工作仍在继续，也不清楚那些离开政府的人员是否还能获得他们曾经依赖的相关信息和联邦人脉。

这位分析师回忆说，大约两年前，他们曾与运营 Facebook 和 Instagram 的 Meta 公司就相关问题开展合作，当时双方保持着良好的工作关系。这位分析师无法确定如今双方的合作是否仍在继续。

同样，一位美国前官员回忆说，2020年大选的筹备工作大约在投票前一年就开始了，随着选举日的临近，协调工作也越来越频繁。有些电话会议有80到100人参加，其中包括一些私营公司，讨论的话题包括选票安全和发布准确的投票信息。

“当私人、私营部门、行业和政府共同努力，朝着保护我们的自由和生活方式这一共同目标努力时，我们国家的安全才能得到最好的保护。这包括举报可疑和非法活动，并利用法律途径和权力来保护美国人民和利益。” Wiley Rein LLP 的特别顾问、前 FBI 高级主管 Erin Joe 在接受 Nextgov/FCW 采访时表示。

在ABA会议上，前国土安全部部长迈克尔·切尔托夫指出，政府在帮助民众识别欺骗方面仍然发挥着作用。官员可以解释信息的来源，或者证明所谓的录音是由人工智能生成的，而无需引导民众接受某种政治观点。

切尔托夫说：“这是用好的言论对抗坏的言论。”

TikTok发言人建议Nextgov/FCW访问一个网页，该网页概述了该公司为应对中期选举和打击平台上的秘密影响力行动而采取的措施。包括Meta、X、Anthropic和OpenAI在内的其他几家社交媒体公司和人工智能公司均未回应置评请求。

两位州选举官员描述了他们的办公室如何应对误导性信息，同时收到的与以往选举周期相比更少的联邦选举简报。

一位人士表示，识别虚假信息的幕后黑手可能很困难。他们的办公室通常重点关注可能混淆民众对投票时间、地点和方式的认知，以及对选举工作人员的威胁。

当出现误导性言论时，公关人员和领导层会决定回应是否有助于选民，或者是否会引发更多关注，导致一些原本可能被忽略的事情最终不了了之。这位官员表示，针对选举官员的威胁将移交执法部门处理。

这位官员表示，他们“已经有一段时间”没有收到专门针对选举的联邦简报了，尽管他们定期参加网络安全和基础设施安全局 (CISA) 召开的关于软件漏洞和已知缺陷被黑客利用的电话会议。这位官员说，这些讨论很少提供与选举相关的具体信息。州情报融合中心每月仍会召开会议，汇集州、地方和联邦执法部门，讨论选举官员面临的威胁。

第二位官员表示，他们的办公室会在联邦选举威胁简报会举行时派代表参加，但这些简报会的举行频率低于 2024 年。

不过，这位官员表示，他们的选举办公室比往年准备得更加充分，制定了多项备用计划，以应对可能发生的干扰。

“我们希望不必用到它们。但如果需要，它们就在那里，”这位官员说。

随着人工智能降低开展影响力营销活动的成本和所需精力，这些准备工作也随之展开。网络安全公司Horizo​​n3.ai的首席执行官兼联合创始人斯内哈尔·安塔尼表示，人工智能在协助开展这些行动中发挥了关键作用，从撰写和分发帖子到研究人们的反应，无所不包。他指出，运营人员可以测试不同的信息，并利用结果调整策略，而人工智能代理可以自动完成大部分工作，从而减少以往所需的时间和专业知识。

安塔尼表示：“人工智能极大地降低了所需的技能，并极大地加快了这一周期的执行速度。”

另一位美国前官员补充说，功能日益强大、可下载的开源模型使规模较小的参与者能够开展以前需要大量投资和专门工程才能完成的工作。

这位前官员表示，“准入门槛要低得多”，并补充说，人工智能生成的文本可以融入日常在线对话中，比有明显缺陷的篡改图像或视频更难被发现。

另一位现任官员表示，俄罗斯本轮周期更有可能采取更为明显的煽动性宣传，而中国的努力则可能更为隐蔽，以服务于更长远的目标。他还补充说，预计两国都将更多地利用人工智能。尤其是在2月底爆发的战争中，亲伊朗组织近几个月来更加公开地利用人工智能。

但这位官员提醒说，更强大的AI工具和更具说服力的内容并不一定意味着外国势力正在改变人们的投票意向。他们补充说，一张图片或一段视频获得大量浏览并不意味着人们会相信它，或者会因此改变投票意向。

在某些情况下，影响力宣传活动甚至可能并非针对符合投票资格的选民。例如，针对年轻美国人的内容可能会影响他们的政治立场，但他们可能尚未达到法定投票年龄。

而在其他情况下，可能并不需要外国势力干预才能让美国公民以某种方式投票。

“如果我看到汽油价格是每加仑 7 美元，我就不一定需要更多理由去投票给民主党了，”这位官员说道，他指的是加州选民，该州部分地区的燃油价格已经创下历史新高。

在这些来自国外的威胁出现之际，国内的政治竞选活动也越来越多地在其广告中使用人工智能。

网络安全公司 Check Point 周三发布的研究显示，人工智能生成的政治视频的日益普及，也可能使人们更难区分真实视频和捏造的视频。

该公司的研究结果显示，“视频是让某种说法显得可信的便捷方式之一，因为人们往往更相信自己亲眼所见、亲耳所闻。”随着人工智能生成的政治内容日益普及，选民越来越难以“相信候选人的视频是真实的”，而任何人也更容易“将真实的视频误认为是伪造的”。

前中央情报局案件官员金伯利·卡尼表示，如今虚假信息的规模和复杂程度“使我们对它的敏感度可能比过去几年要低一些”。

“作为个体，我们必须渴望了解我们所接触的信息是真是假，或者它是否迎合了某种对我们有利或更舒适的叙事，”她在一次采访中说道。“所以，一切都始于认知上的诚实和觉察，想要知道你所接触的信息是否准确，或者是否存在偏见。”

下一篇报道：美国陆军将在南美洲举行禁毒训练演习