Rethinking organizational design in the age of agentic AI
Amid rapidly growing adoption of enterprise-level AI agents, there’s a disconnect emerging between ambition and execution.
Although 85% of organizations say they want to be agentic within the next three years, 76% say their current operations and infrastructure can’t support that change. They cite a lack of readiness across people, processes, and workflows.
The sticky tape problem
The challenge is that many organisations are often layering AI agents onto existing operations, rather than reimagine the operating model and how work will need to be rewired, explains Prasun Shah, global CTO for workforce consulting and chief AI officer at PwC UK Consulting. “They’re embedding AI employees into what is a human operating model,” layering on AI agents to existing workplace structures when “this is like adding sticky tapes to parts of an operating model that is breaking.”
Doing so may be preventing organizations from unlocking the full value agentic AI offers, creating circumstances where disillusionment can quickly creep in. That full value lies in agents’ capacity to execute entire workflows with limited human input. They can coordinate complex tasks, make independent decisions, adjust to changing conditions, and iterate performance.
In early proving grounds that span customer service, HR, and sales, it’s already estimated that AI agents could accelerate business processes by as much as 30% to 50% and low-value work time by 25% to 40% when deployed at scale. But with this capability comes greater complexity and the need for an enterprise-wide change.

Growing the AI vocabulary
Enterprise agentic AI platform Ema describes this change as agentic business transformation (ABT), a term it coined last year in partnership with HFS Research, in an attempt to plug what it sees as a gap in the existing lexicon about AI agents, and to provide enterprises with a new framework with which to think about their own adoption of the technology.
“None of the existing vocabulary captures the full scope of the change,” explains Ema CEO and founder Surojit Chatterjee. “Digital transformation was about moving from paper to software. AI transformation was about adding artificial intelligence to existing processes. Co-pilot is about AI assisting in various human tasks. But ABT is something categorically different: It’s the integration of AI agents into the fabric of the organization.”
For Shah, the dedicated term (ABT) “helps drive the need to redesign an organization in its entirety: its operating model, its workflows, decision rights, and performance management systems.” He emphasizes that “everything that’s needed to ensure those agents are actually active participants in value creation, rather than just point tools or productivity aids.”
According to Ema, ABT encompasses three core pillars: an organization’s technology stack, its workforce, and the metrics used for success.
AI agents as connective tissue
The first pillar of ABT is the technology stack. “Your existing tech stack was designed for human-operated, application-centric workflows,” says Chatterjee. “It needs to be reconsidered when the actor is an AI agent operating at machine speed across multiple systems simultaneously.”
As AI agents are integrated into an organization, enterprises will need to pivot from a set of linear processes and steps, to rewiring work in a very different way, explains Shah. That’s because the value in AI agents isn’t as another layer in an existing technology stack but as a connective tissue, he explains, moving between or across layers to coordinate a high-level task or retrieve and interpret data from multiple discrete applications. AI agents can create “a true competitive differentiation for an enterprise” by making decisions based on this capacity to contextualize, he says. “That is where the next battleground will be.”
To build this connective tissue, leaders need to adapt their technology stack to surface higher quality decisions from AI agents, prioritizing access to multiple datasets and applications simultaneously to develop tacit knowledge. “Organizations that make this architectural shift become genuinely more adaptive,” says Chatterjee. “When a new business requirement emerges, you don’t wait six months for a software vendor to build a feature. You configure an AI employee using natural language and connect it to the systems it needs. The time from business to production workflow drops from months to days.”
The workforce, redesigned
As AI agents are deployed for more use cases, enterprise leaders must consider what this means for dynamics across their workforce, the second pillar of ABT.
Workforce structures today deviate little from the hierarchical model of the early days of industrialization. To maximize efficiency and scale, processes are standardized, tasks are clearly delineated between strategic business units (SBUs), and employees progress up through an organization based on their capacity to optimize output from teams below them. But with AI agents that can execute, coordinate, and optimize tasks—often without managerial coordination—the lines of that established hierarchy become blurred.
In a workforce that blends AI agents and human employees, managers will be freed up from many execution-based tasks but take on new responsibilities associated with managing hybrid teams. Managers “will need to be able to manage issues around trust, explainability, psychological safety, and even status dynamics” to navigate new tensions that could arise in a hybrid workforce, says Shah.
The impact of agentic AI on existing workforce structures goes far beyond the management layer, too. McKinsey predicts that by 2030, three-quarters of current jobs will require redesign, upskilling, or redeployment, and organizations will need to act swiftly to amend recruitment, retention, and remuneration.
From output to outcome
Success metrics are the third and final pillar of ABT.
As AI agents assume greater ownership of core enterprise processes, taking on collaborative roles alongside human employees, traditional workforce metrics that focus on activity or output—such as calls handled or reports filed—no longer make sense.
“When you add AI employees into the workforce, activity metrics become meaningless or actively misleading,” says Chatterjee. “An AI employee can handle a thousand customer interactions in the time it takes a human to handle ten. If you measure success by interactions handled, you’ll conclude the AI is working brilliantly while missing whether any of those interactions actually drove customer satisfaction, retention, or revenue.” To correct this, enterprises must develop a new set of metrics that focus on outcome rather than output. That is, metrics on the broader benefits or changes achieved, rather than individual deliverables.
For example, when one of Ema’s large enterprise customers overhauled its own metrics, switching from tool metrics like cost per query and AI accuracy, to outcomes like the percentage of contracts reviewed without human escalation, the measured ROI from agentic AI tripled within two quarters. The changes meant “this customer stopped building point solutions in high-volume, low-complexity workflows and started deploying AI employees where the outcome value was highest,” says Chatterjee.
Integrating new metrics may also require a complete reconfiguration of reward and talent management processes, as well as accountability and ownership within organizations, points out Shah. In human-AI teams, for example, although ethical and fiduciary responsibilities will likely remain with human employees, operational accountability will become significantly more diffused to reflect the systemic role of AI agents.
This change will raise new questions that senior leadership teams will need to wrestle with, Shah adds. They’ll need to consider: Who is accountable when an AI employee makes a mistake? What happens when AI and humans disagree? What guardrails should be erected to safeguard customers?
Laying the groundwork for systems-level change
Systems-level change is gradual. These are complex lines of inquiry that experts continue to grapple with. But in kickstarting internal dialogue about the core pillars of ABT—the workforce, the technology stack, and the metrics by which success can be gauged—leaders can lay the groundwork for an enterprise better poised to embrace AI agents at a systems level and start to close the gap between their ambition and execution.
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
The impact of face-to-face social exclusion on university students’ interpersonal cooperation behavior: a hyperscanning study based on fNIRS
Synaptic alterations are preceding the axonal loss in optic atrophy of Wolfram syndrome mouse model
Rotherham NHS FT cuts IT help desk calls by 28% with AI
Using Illusions to Boost Social Cognition
Interventions: Behavioral: Visual-illusion social cognition session (immediate); Behavioral: Visual-illusion social cognition session (delayed)
Sponsors: Hôpital le Vinatier; INSERM 1028
Not yet recruiting
Scaling creativity in the age of AI
Storytelling is core to humanity’s DNA, stemming from our impulse to express ideals, warnings, hopes, and experiences. Technology has always been woven through the medium and the distribution: from early humans’ innovation of natural pigments and charcoals for cave paintings to literal representation by the camera.

The landscape of storytelling continues to shift under our feet. Social and streaming platforms have multiplied, audiences have fragmented, and our demand for fresh, unique media is insatiable. A recent McKinsey podcast cites that we are watching upwards of 12 hours of video content daily, often on multiple devices and multiple platforms.
All this content is expensive to produce: With a baseline budget of $150M, a Hollywood feature runs $1M per minute of finished film; prestige streaming content is in the hundreds of thousands per minute. And since consumers want to engage with authentic, original material, every company is now effectively a media company. That means we all face the same pressure: more content, with the same time and budget constraints.
There is no longer a question whether to use AI for content; the math doesn’t work any other way. What leaders need to focus on now is how to adapt responsibly, protect brand integrity, uplift team creativity, and build customer trust.
A few things worth holding onto as this era accelerates:
- AI amplifies what’s already there, both good and bad. Weak strategy stays weak.
- Responsible adoption means knowing what’s in your tools and models. Provenance and transparency are the foundation, not the finish line.
- Scale without taste is just noise. Investing in your team’s judgment is what makes more content matter.
- Fundamentals of great storytelling have not changed. Regardless of format or channel, what makes audiences lean in are still characters, arc, ingenuity, and surprise.
The permanent sprint
Creative teams are trapped on the endless hamster wheel of production, and it’s not slowing down. According to Adobe research, content demand will grow 5x over the next two years. Social content shelf life is now measured in hours, not weeks. Keeping fresh work in the pipeline is a permanent sprint, requiring teams to rethink how creative production functions.
The first move is freeing creative teams by having AI absorb the repetitive work so they have space for the strategic creative decisions that require human ingenuity. In a recent study from Adobe, 94% of creatives report that AI helps them produce content faster, saving an average of 17 hours per week. That recovered time is not a productivity metric; it is renewed creative capacity.
As a use case, Nestlé offers a useful blueprint. Its teams operate across 180 countries with a portfolio of iconic brands including Nescafé, KitKat, and Purina. Using Adobe Firefly Custom Models embedded in existing content workflows allows teams to generate assets in a brand-informed style without disrupting creative flow. At Nestlé, workflow cycle times dropped 50%. “With Firefly Custom Models, we can react at the speed of culture. It’s the closest thing we’ve had to magic.” says Wael Jabi, global strategic comms lead for KitKat.
As we move into the agentic era, the possibilities expand further. Adobe’s Creative Agent thinks in systems, not tasks, orchestrating across workflows, apps, and processes to close the gap between idea and execution, and get teams out of the production cycles that consume their productivity.
Build for your brand, not every brand
A company’s brand is how the world recognizes and connects with them. And it’s more than a collection of assets—it is dynamic, subjective, and expressed in thousands of micro-decisions made every day by the people who know it best. As production scales, keeping everything tuned to the brand gets more challenging. Off-the-shelf AI cannot replicate the level of nuance creative teams bring to content, and there’s a real cost to getting it wrong; diluting a brand in market with almost-right output is not an acceptable option. Customer trust is fragile.
Starting with a bespoke AI model built with Adobe Firefly Foundry addresses this directly. Firefly Foundry starts with a commercially safe base model and trains further on a company’s IP, making it possible to produce content that genuinely reflects the team’s vision.
And to ensure that Firefly Foundry models truly represent the creatives at the helm, Adobe has partnered with film studios like Wonder Studios, Promise.ai, and B5 Studios, and the “big three” talent agencies CAA, UTA, and WME to deeply understand what it means (and what it takes) to build an IP-immersive model that keeps creatives at the center as these film studios and talent agencies scale their visions. These brand ecosystems can accelerate nearly every phase of the production process, from ideation and storyboarding to production and promotion, all while preserving artistry and authorship. And to power the next generation of creativity and content, Adobe has recently announced a strategic partnership with NVIDIA, delivering best-in-class creative control along with enterprise-grade, commercially safe content at scale.
Generic AI gives teams a starting point. But a model trained on a brand’s own IP gets them to the finish line, while still leaving room for the creative calls that matter most.
When agents become the audience
AI is not only reshaping how we create; it is reshaping how customers find and engage with brands entirely. According to Adobe Digital Insights, AI-powered shopping has surged 4,700%. Agentic web traffic is up 7,851% year over year. Yet, most businesses still have significant gaps in AI-led brand visibility. If content is invisible to AI agents, then a brand is invisible to customers.
Major League Baseball is ahead of this curve. Using Adobe LLM Optimizer, the league monitors how its content surfaces across AI interfaces and makes real-time adjustments to maintain visibility. As fans search for tickets, stats, or game-day experiences, the league ensures its brand shows up wherever that search is happening. And with Adobe’s recent acquisition of Semrush, brand visibility goes even further.
The agentic web created an entirely new content surface that did not exist two years ago, and this exponential proliferation of content illustrates precisely why scaled, on-brand content production has become a strategic imperative. A well-built agentic foundation offers full visibility into (and control over) every piece of content, from production to performance.
How to prepare for AI integration
Here are a few steps to get started:
Audit before automation. Content supply chains usually include duplicated processes, unclear ownership, and assets living in many different places. Before AI can accelerate anything, develop a clear map of how content moves through the organization today: who creates it, who approves it, where it lives, and where it breaks down. AI applied to a broken process just breaks it faster.
Walk through workflows. Resist the urge to overhaul everything at once. Start with production tasks that are high-volume, low-stakes, and well-defined: asset resizing, localization, and background generation. Use those wins to build internal confidence before expanding into more complex creative territory.
Build responsible governance from the start. Governance added as an afterthought becomes a bottleneck. Building it in from the beginning creates a competitive advantage that lets teams move fast with confidence. And this means clear policies on model training, content provenance, human review thresholds, and communicating AI use to customers. The brands that earn lasting trust will treat transparency as a feature, not a footnote.
This content was produced by Adobe. It was not written by MIT Technology Review’s editorial staff.
Oncology’s Next AI Battleground: Instant Clinical and Commercial Insight
Across the oncology pharmaceutical industry, the bar for precision is constantly being raised. Cancer drug development has become increasingly biomarker-driven, trial populations are narrowing, and the cost of identifying eligible patients for studies continues to rise. At the same time, life sciences organizations are under growing pressure to generate real-world evidence (RWE) faster for commercialization strategies as well as regulators and payers.
The race to operationalize artificial intelligence (AI) across oncology research has entered a new phase. After years of building massive catalogs of real-world data (RWD) from electronic health records (EHRs), molecular testing, and longitudinal patient outcomes, healthcare technology companies are now competing to transform those datasets into interactive intelligence systems capable of answering complex clinical and commercial questions in real time.
That convergence has fueled a wave of oncology AI platform development from companies including SOPHiA GENETICS, Ontada, COTA Healthcare, and now Flatiron Health. “As oncology becomes more complex, the ability to quickly identify the right patients and answer critical research questions is no longer a nice-to-have, it’s essential,” Kate Estep, chief product officer at Flatiron Health, told Inside Precision Medicine.
The move by Flatiron Health supports the continuing trend of data companies in oncology and across healthcare positioning themselves beyond simply aggregating clinical datasets toward creating AI-native research environments where clinicians, commercial strategists, and researchers can interact directly with data using natural language.
The need for speed
Historically, RWE generation has been labor-intensive. Pharmaceutical teams often rely on analysts or biostatistics groups to construct cohorts, validate inclusion criteria, and generate feasibility assessments—a process that can take days or weeks before a research question even begins to take shape. That workflow is increasingly incompatible with modern oncology development, where therapies are often targeted toward highly specific molecular subpopulations.
Cancer research may be uniquely suited for AI-native evidence generation systems. Compared with many therapeutic areas, oncology already produces unusually data-dense patient journeys involving pathology reports, genomic sequencing, imaging, biomarker testing, treatment lines, progression tracking, and survival endpoints. Oncology drug development is also increasingly dependent on identifying narrow molecular populations quickly and accurately. That complexity creates ideal conditions for conversational AI systems capable of navigating structured and unstructured clinical data simultaneously.
Flatiron Telescope attempts to address that bottleneck by giving users a conversational interface layered on top of Flatiron’s oncology-specific datasets. Researchers can describe inclusion and exclusion criteria in natural language, dynamically refine cohorts, and immediately view patient counts, attrition curves, treatment patterns, and survival analyses without writing code. “We were chatting with one of our early access partners last week, and this person was remarking, ‘I could answer my question in 30 minutes, and that would have taken me two days before waiting for my data team to come back to me,’” Estep said during a media briefing ahead of launch.
That acceleration may ultimately become the defining metric in the AI healthcare infrastructure market: not simply the size of a company’s dataset, but how quickly actionable insight can be extracted from it.
From data vendors to research platforms
But the challenge is not merely access to information. Trust and scientific validity remain central concerns. “One of the things our head of data science was sharing is that off-the-shelf models are roughly 60% accurate,” Estep said. “When built and trained with the clinical and scientific best practices that we have applied to model context because we have been asking cohort questions of our data for 15 years, that’s 90% plus accuracy.”
Those comments point toward an increasingly important divide in healthcare AI: the distinction between general-purpose AI models and clinically fine-tuned systems trained on domain-specific workflows. For companies like Flatiron, the competitive moat may ultimately come less from the underlying language models themselves and more from proprietary clinical context, curation methodologies, and validated evidence-generation pipelines.
The emergence of platforms like Telescope also reflects a broader transformation occurring across healthcare AI. The first generation of healthcare data companies focused primarily on aggregation, assembling electronic health record (EHR) data, claims data, genomic profiles, and imaging repositories into structured datasets. The second generation is now focused on orchestration: enabling users to interrogate those datasets continuously through AI-driven systems.
Flatiron is betting that domain specificity will matter more than generic AI capability. “Most people in the space either give you data, they give you analytics, or they give you a platform,” Estep said. “Very few cut across all three buckets.”
That positioning distinguishes Flatiron somewhat from competitors. Tempus AI has built a broad precision medicine tech ecosystem for both providers and life sciences companies. SOPHiA GENETICS has emphasized multimodal analytics and genomic interpretation. Ontada, backed by McKesson, combines oncology data assets with point-of-care tools and network analytics.
Flatiron, by contrast, is leaning heavily into its reputation for longitudinal oncology RWE and EHR-derived clinical depth. The company says Telescope is powered by more than 15 years of oncology-specific data infrastructure spanning over 4,700 providers and 1,600 clinical sites in the United States, representing approximately 40% of U.S. community oncology practices. Globally, the company now manages data from more than five million patient journeys across the U.S., U.K., Germany, and Japan.
That scale matters because oncology AI systems depend heavily on context-rich longitudinal data. Large language models alone are insufficient if the underlying clinical infrastructure lacks standardized outcomes, biomarker histories, treatment sequences, or progression events. “Flatiron has spent the last decade and a half building high-quality, oncology-specific real-world datasets,” Estep said. “Telescope really sits at the epicenter of that.”
Global oncology intelligence
Another major shift underway in oncology AI involves international interoperability. Historically, most RWE systems were fragmented geographically, with datasets built independently for different markets. But as pharmaceutical companies globalize clinical development programs, pressure is increasing to harmonize datasets across countries.
Flatiron says it is now building globally interoperable oncology datasets across the U.S., U.K., Germany, and Japan, beginning with prostate cancer data expected later this year. “We are ensuring that our datasets are interoperable from a global perspective,” Estep said. “Conclusions drawn on definitions of variables and data models in one market can easily be applied or explored in another.”
The long-term implications could be substantial. Globally harmonized oncology datasets would allow researchers to study treatment variation, biomarker prevalence, and outcomes across healthcare systems at a scale previously difficult to achieve. It may also help address longstanding concerns around representativeness in RWE generation. “Representativeness of a RWD is probably one of the single biggest requirements for us as we think about whether this dataset is considered reliable,” Estep said.
Perhaps the most important industry trend underlying Telescope’s launch is the democratization of advanced analytics. Historically, sophisticated oncology data analysis required teams of data scientists, epidemiologists, or biostatisticians. AI interfaces are beginning to collapse those barriers, enabling clinical operations leaders, medical affairs teams, and commercial strategists to interact directly with research-grade datasets. “There’s no coding required,” Estep said. “Any team member can use it, not just your data analysts.”
That shift could fundamentally change how oncology organizations make decisions, reducing delays between hypothesis generation and evidence generation while broadening access to sophisticated analytical capabilities across enterprise teams.
Whether Telescope ultimately becomes a dominant platform remains to be seen. But its launch reflects a broader reality now reshaping healthcare: in oncology, the future competitive advantage may belong not simply to companies with the most data but to those capable of turning clinical complexity into usable intelligence fastest.
The post Oncology’s Next AI Battleground: Instant Clinical and Commercial Insight appeared first on Inside Precision Medicine.
Tech researchers are suing the Trump administration over the future of online safety
Since its earliest days back in office, the Trump administration has been going after researchers who study and try to counter hate speech, harassment, propaganda, and disinformation online.
Now, some of those researchers are fighting back. Last week their lawsuit—which could have global repercussions for online safety and free speech—made its first appearance in court.
This fight started a year ago, when US Secretary of State Marco Rubio announced on X what he called a “visa restriction policy” against “foreign officials and other persons” who were “complicit in censoring Americans.” Since then, a handful of foreign officials and researchers have been barred from travel to the US, and in theory, anyone working in fact-checking or online trust and safety more broadly could face the same restrictions.
Still, the exact implications of Rubio’s announcement are unclear—purposefully so, argues Carrie DeCell, a lawyer representing the researchers. “This policy is expansive and incredibly vague, and the chilling effects are correspondingly enormous,” DeCell said outside the courthouse in Washington, DC, on May 13.
The case has been brought by the Coalition for Independent Technology Research (CITR), an advocacy organization for tech researchers. It is suing Rubio, former US secretary of homeland security Kristi Noem, and former US attorney general Pam Bondi and asking the court to strike down the policy as unconstitutional. In their complaint, the plaintiffs say the policy violates the speech and due process rights of foreign-born tech researchers and workers whose “work supports greater moderation of content on the [tech] platforms.”
CITR is represented by Columbia University’s Knight First Amendment Institute and the legal nonprofit Protect Democracy. DeCell, a senior staff attorney at the Knight Institute, tells MIT Technology Review that they’re in court because the Trump administration is effectively “using immigration law to punish people for expressing views that it disagrees with.”
This story is part of MIT Technology Review’s “America Undone” series, examining how the foundations of US success in science and innovation are currently under threat. You can read the rest here.
Most immediately, the plaintiffs are asking the government to halt these visa restrictions while the case proceeds. Zachariah Lindsey, the assistant US attorney representing Rubio and the other defendants, argued in last week’s hearing that the government is not targeting speech but, rather, “conduct [that] is assisting or facilitating foreign government censorship of free speech.” At the end of the week, the government filed a motion to dismiss the case.
The judge has yet to rule on either motion, and his questions so far appeared to focus on parsing what (and who) is actually affected by the State Department’s announcements, as well as other procedural issues.
The outcome of the case may ultimately affect how much the public knows about the risks of social media and AI, says Nicole Schneidman, head of Protect Democracy’s technology and data governance team. The workers bringing this suit, she says, “serve a really, really important function in educating the public, holding tech companies accountable, doing research on the ramifications that advanced technology has on our society.”
“A political witch hunt”
CITR’s lawsuit is the latest salvo in a yearslong battle over how the internet should be moderated, and by whom—a question that has become increasingly political and entangled in allegations of censorship.
For years, Trump and his allies have claimed to be victims of a vast conspiracy between government agencies, civil society groups, academics, and Big Tech platforms to specifically censor conservative voices online. According to this narrative, a so-called “censorship-industrial complex” helped the Biden administration subvert First Amendment protections on speech by allegedly outsourcing censorship to these groups.
The State Department claims Rubio was able to implement the immigration policy because the Immigration and Nationality Act authorizes him to “render inadmissible any alien whose entry into the United States ‘would have potentially serious adverse foreign policy consequences for the United States.’” Before the current Trump administration, the statute was rarely invoked, and when it was, it was typically with more limited, specific criteria, rather than its current application against anyone who has participated in alleged censorship—an action that has no legal definition.
The administration first deployed the policy in July 2025, when Rubio issued a statement announcing the revocation of visas for Alexandre de Moraes, the lead justice on the Brazilian Supreme Federal Court, and “his allies on the court” who were involved in prosecuting Jair Bolsonaro, Brazil’s former president. The prosecution was a “political witch hunt,” said Rubio, calling it evidence of a “censorship complex so sweeping that it not only violates basic rights of Brazilians, but also … targets Americans.”
Then, in early December, the State Department issued instructions to embassies to reject H-1B visa applications from individuals who had worked specifically in fact-checking, online trust and safety, and mis- or disinformation research, as Reuters first reported.
A few weeks later, on December 23, the agency announced visa restrictions for five Europeans whom it accused of censoring Americans. This included two CITR members: Imran Ahmed, founder and CEO of the Center for Countering Digital Hate, which documents hate speech on social media platforms, and Clare Melford, cofounder of the Global Disinformation Index, which ranks websites according to how often they publish hate speech and disinformation. Also banned were the former European Union commissioner Thierry Breton, a key architect of the European Union’s Digital Services Act (which the State Department has called “Orwellian” and an example of censorship), and Josephine Ballon and Anna-Lena von Hodenberg, co-CEOs of HateAid, a German nonprofit that fights online hate speech.
Ahmed, who lives in the US with his American wife and child, quickly filed his own lawsuit to stave off deportation and halt the policy. A preliminary injunction preventing his detention and deportation is in place as the lawsuit continues.
The Department of Homeland Security referred questions from MIT Technology Review to the State Department, which referred “specific questions” to the Department of Justice, while also writing that “the Trump Administration believes that aliens who are or were involved or complicit in censoring American citizens must face appropriate consequences. An American visa is a privilege not a right.” The Department of Justice did not respond to a request for comment.
“A gut punch”
Now, more tech researchers are fighting back.
CITR represents 500 individual and institutional members in 47 countries; 40 are based in the United States, including around 30 noncitizens. The organization argues that US-based tech researchers are experiencing a widespread chilling effect and are having to change or reframe what they’re studying so that it’s less explicitly (or less obviously) about content moderation or countering disinformation. Alternatively, some are leaving the US altogether, or making plans to do so, in order to safely carry out their work.
CITR member Eirliani Abdul Rahman, a Singaporean online safety expert and a founding member of Twitter’s Trust and Safety Council, is one of these individuals. Her experience was included, though described anonymously, in CITR’s initial legal complaint.
Back in December 2022, shortly after Elon Musk purchased Twitter, Rahman and two other Trust and Safety Council members publicly resigned. They spoke out against “red lines” the new owner had crossed, including his reinstatement of accounts that had previously been banned, and noted the marked increase in hate speech on the platform.
Musk disbanded the council days later, but first he retweeted a post that tagged Rahman and the others and said: “You all belong in jail.” This led to a level of online harassment, doxxing, and death threats that she had never before experienced. “I was trained as an economist, and I could just see line graphs form in my head of the stochastic jump in what happened,” Rahman says, referring to the way the dangerous attention spiked after Musk effectively endorsed the other user’s provocation.
This experience inspired her to pursue a new area of research: using quantitative methods to study and hopefully stop social media harassment “in real time,” she says.
“The ones that are most harassed are people [who] have historically been marginalized,” she adds. “Most of us know about this already, like it’s intuitive. But until you quantify it, sometimes it’s just not seen and taken seriously.”
But then Trump was reelected, making the work feel untenable. The US quickly became “a funding desert” for scientific research, she says, with federal support for any research perceived by conservatives to focus on mis/disinformation getting cut. At the same time, tech companies shifted their positions on content moderation to align with the president’s, meaning that her research would be unlikely to have any practical implications: “There’s simply no guardrails around social media anymore,” she says.
Fast-forward to December 2025, and the travel bans on the five Europeans felt like “a gut punch to the stomach,” Rahman says. She and Ahmed had both testified earlier in the year before the UK Parliament on the role social media played in spreading false claims about the supposed Muslim identity of a murderer who had killed three British girls; this online activity contributed to violent anti-immigrant and Islamophobic riots across the country in the summer of 2024.
The targeting of Ahmed and the other Europeans “was the last straw” for Rahman. Soon after, she left the US for a three-year fellowship in Germany aimed at supporting “international academic freedom”—coincidentally arriving in the country on the same day CITR filed its lawsuit.
“My body just calmed down,” Rahman says of landing in Germany. “I didn’t wake up in the middle of the night … always wondering about the next executive order and how it pertained to my situation.”
Rahman believes this legal battle has implications that reach beyond CITR members and their families. It “pertains to all immigrants in the US to protect our First Amendment rights,” she says.
Additionally, whether fact-checkers, online trust and safety workers, and tech researchers can continue to do their work has a broader impact on anyone who uses the internet.
Earlier this year, for example, Ahmed’s Center for Countering Digital Hate published widely cited research that Grok’s image-editing feature had generated an estimated 3 million sexualized images, including 23,000 images of children, in an 11-day period. This led to government investigations, lawsuits, and even temporary bans for Grok’s parent company, xAI, across the United States and world.
“The threats have really sharpened”
MIT Technology Review has reported extensively on this right-wing war on supposed censorship; one of our stories revealing that State Department leadership requested communications records from a now-shuttered office focused on countering foreign disinformation has been included as an exhibit in the CITR lawsuit. This request sought insight into communications with a slew of individuals some far-right activists allege are involved in the “censorship-industrial complex,” including journalists, the German foreign minister, and numerous researchers studying disinformation and hate speech (including Medford, Ahmed, and their organizations).
DeCell tells us that over the past year and a half, there have been more lawsuits against the Trump administration regarding free speech—because “the threats have really sharpened,” she says.
Last year, the Knight Institute sued Rubio on behalf of of university faculty and students who have been arrested, detained, and deported for their pro-Palestinian speech; this past January, a judge ruled that the administration’s deportation policy was unconstitutional. The risk to free speech rights is “palpable” when the government “decides to target people specifically with the threat of rounding them off the streets, throwing them into a detention center, and then potentially deporting them from this country,” DeCell says.
Though Rahman is safely abroad for now, she says she’s watching the CITR lawsuit closely. Ultimately, she says, she believes it will determine whether researchers will be able to continue to do their work, “which is to take social media platforms to account,” she says—“making sure there’s actual accountability and independent oversight is critical to protecting our democracies.”

