Category AI News

Whether you’re playing poker against a single opponent or find yourself in a bidding war over a home purchase with another prospective buyer, you are operating under conditions of imperfect information. You know what cards you’re holding in the poker game, and you also know how much above the home’s asking price you can afford, but you don’t know your opponent’s hand in the card game or how high the other home buyer is willing to go. 

A paper co-authored by MIT researchers and presented in April at the International Conference on Learning Representations in Rio De Janeiro won’t tell you what to do in these situations, specifically. But it does offer new insights into so-called imperfect-information games that involve two contestants facing off in a “zero-sum” competition, where one player’s gain means the other player’s loss.

MIT researchers on the project include Sobhan Mohammadpour, a PhD student in MIT’s Department of Electrical Engineering and Computer Science (EECS) and the Laboratory for Information and Decision Systems (LIDS); and Gabriele Farina, an assistant professor in EECS and a principal investigator at LIDS. Additional co-authors include Max Rudolph of the University of Texas at Austin (UT), Nathan Lichtlé of the University of California at Berkeley (UCB), Alexandre Bayen of UCB, J. Zico Kolter of Carnegie Mellon University (CMU), Amy X. Zhang ’11, MNG ’12 of UT; Eugene Vinitsky of New York University; and Samuel Sokota of CMU. 

The focus of the new work is on algorithms that could be used to train neural networks to participate in imperfect-information games. The assumption, long-held in the field, was that algorithms grounded in principles of game theory would, in this setting, clearly outcompete a general-purpose variety of algorithms called policy gradient methods, which came into use for decision-making in the 1990s. The term “policy” in this context basically means strategy, whereas “gradient” refers to a path that leads in the direction of greatest change — to the top (or bottom) of a hill, for example. Policy gradient methods are being used to train neural networks to make decisions that move — in small, sequential steps — toward a particular goal (like reaching a summit, metaphorically speaking), with continual adjustments and course corrections made along the way to bring the agent closer to the intended destination.

Although strategic games were not on the original agenda when policy gradient methods were conceived in the early 1990s, the authors of the new paper still wondered how this class of algorithms might fare in two-player games. These methods become more complicated to analyze in multi-agent settings, according to Farina. “There is still a direction you can move in to improve your circumstances, but, because of the other player’s actions, that direction can constantly change over the course of the game. And those shifts can be rapid.”

“It had been pretty much taken for granted that specialized game-theoretic algorithms were the right approach for this setting,” says Sokota. “Our study showed that policy gradient methods can work better than these specialized algorithms, and that the specialized algorithms may not work as well as people thought — which raises an interesting sociological question about why this went unnoticed for so long. Part of the answer is that the field hadn’t done the engineering work required to rigorously evaluate the algorithms, so it was hard to tell what worked and what didn’t.”

Consequently, a major contribution of this work has been to provide an even-handed way of appraising different algorithms that can teach agents — i.e., neural networks — how to compete in imperfect-information games. “We’re taking a different approach,” notes Rudolph. “Unlike many of the papers published in this field, we’re not proposing a new algorithm that can beat out other algorithms. We’re proposing a benchmark that can assess these algorithms.”

Simply put, a benchmark consists of software designed to rate the performance of algorithms. “What we’re offering is a testing grounds, or playing grounds, where people can take their algorithms, train them for a specific task, and see how well they do,” says Farina.

The group calculates a player’s performance in terms of a concept called exploitability, which measures how well a player does against the “worst-case adversary,” Sokota explains. “In a game like poker, this opponent wouldn’t know what my hand is, but would know how I would behave for any given hand.” Achieving a zero on this scale implies perfect play, whereas a high exploitability score indicates far-from-optimal play.

Five games were played in experiments carried out by the team: two versions of Phantom Tic-Tac-Toe, in which players can’t see what their opponent has done, along with two imperfect-information variants of a board game called Hex, and another game of deception called Liar’s Dice.

The biggest challenge faced by the researchers was getting the exploitability measure to work on games of this size, which may include as many as 30 billion states. A “state” in this case is not just all the possible board positions, but also encompasses the entire history of the game, including every step and misstep along the way. 

“It’s like looking into a dark room that’s filled with objects you can’t see,” says Mohammadpour. “Somehow, you need to figure out where these objects are and exactly how they got there.” Previous researchers, Mohammadpour adds, have typically used exploitability for games that are 100,000 times smaller than the ones analyzed in their study.

In the experiments carried out on these five games, neural networks trained with policy gradient algorithms got better (lower) exploitability scores than networks trained on game theory-based algorithms. In head-to-head competitions, which took place in the next round, the policy gradient-trained networks again beat their game theory-trained opponents. “Those results were reassuring,” Rudolph says, “because they give us more confidence in our benchmarking approach.”

The team has made their benchmarking software freely available and convenient to use. “You don’t need a supercomputer,” Mohammadpour says. “You can run it on an ordinary laptop. And all you have to do is add a single line of code to a commonly used collection of benchmarking software called OpenSpiel.”

Although their experiments involved some fairly obscure games, Farina would like to put this work into a broader context. “Keep in mind that the term ‘game’ really applies to any multi-agent strategic interaction,” he says. “So the lessons we learn from this research are by no means limited to recreational games.”

Vinitsky agrees. “Hidden information is a very important property of the world,” he says. “It pervades a range of things — including military operations, trading scenarios, and negotiations — all of which are carried out under conditions of hidden information. The idea that we can improve on these games suggests that we can also do better in these other settings as well.”

Ian Gemp — a computer scientist and game theory expert at Google DeepMind who was not involved in this study — finds these results encouraging. “This work serves as a compelling reminder,” he says, “that modernizing classical tools [like policy gradient methods] remains a highly productive path for solving complex strategic problems.”

There is a trap buried inside every enterprise AI deployment, and the more useful the agent, the deeper you fall into it. 

A paper published in April 2026 by researchers from Microsoft and Huazhong University of Science and Technology has put a number on the problem and for any AI leader currently scaling agents across their organization, the findings are worth a careful read.

The paper introduces a benchmark purpose-built for the messy reality of enterprise AI: multiple departments, entangled data, hierarchical access rules, and users who sometimes push agents beyond what they should answer. In other words, a fairly ordinary Tuesday inside a large company.

What the researchers found should give pause to anyone currently in the “deploy first, govern later” phase…

The core finding: more capable means more leaky

Across a battery of tests covering GPT-4o, GPT-5, Grok-3, Qwen-2.5, Kimi-K2, DeepSeek-V3, and DeepSeek-R1, privacy violation rates ranged from 15.8% to 50.9%, with information leakage reaching as high as 26.7%.

Those are production-grade models running on realistic enterprise scenarios, failing to keep sensitive information in the right context roughly one in five times at best and one in two times at worst.

The counterintuitive part: higher task utility consistently correlated with higher privacy violations. Agents that were better at completing tasks were also better at pulling in contextual information they had access to, including information they should have withheld.

6 things every AI leader needs to get right in H2 2026

The pilot phase is over. Here are the 6 trends shaping AI strategy in H2 2026, from agentic infrastructure to physical AI and custom builds.

AI Accelerator InstituteAndrew Lovell

What contextual integrity actually means in practice

The theoretical framework the paper uses comes from philosopher Helen Nissenbaum’s concept of contextual integrity: the idea that privacy is violated when information flows to recipients in contexts where it does not belong, even if that information was shared willingly in another context.

An employee sharing health information with HR has a reasonable expectation that a manager asking about team productivity metrics later will be kept away from it. The information was entirely accessible in one context. The context made it private.

Enterprise LLM agents break this constantly. They have access to emails, meeting transcripts, HR records, financial data, and CRM notes simultaneously.

When a user asks a question that touches multiple data sources, the agent has to make a fine-grained judgment about what to include and what to withhold. CI-

Work tests exactly this judgment across five organizational directions:

• Upward flows (employee to manager): whether agents correctly handle information shared with someone more senior in the hierarchy
• Downward flows (manager to team): whether agents appropriately limit what gets shared below the sender’s level
• Lateral flows (peer to peer): whether agents respect boundaries between colleagues at the same level in different functions
• Diagonal flows: cross-functional, cross-level information sharing, where the norms are least clearly defined
• External flows: data shared with parties outside the organization, where the stakes of leakage are highest

The benchmark found that models grasp high-level organizational boundaries reasonably well.

The failures concentrate in the fine-grained cases, specifically where the information is technically accessible but contextually inappropriate to share.

Scaling past the problem can make it worse

This is the finding that carries the most weight for AI decision-makers.

The researchers describe an “inverse scaling” phenomenon: larger models, with greater reasoning depth, sometimes exacerbate leakage rather than reducing it.

The implication is direct: buying a more powerful model is a reasonable response to many enterprise AI challenges. It is a poor response to contextual integrity failures.

The paper’s conclusion is that addressing this requires a shift from model-centric scaling toward context-centric architectures, where the architecture itself enforces what data flows where, rather than relying on the model’s in-context judgment.

Benchmark theater, explained: AI test scores vs production

Every frontier model now scores above 88% on MMLU. So why does a 37% gap still exist between lab benchmark scores and real-world AI deployment performance? We explain why the tests keep lying, and what rigorous evaluation actually looks like.

AI Accelerator InstituteAndrew Lovell

Where agent pressure compounds the risk

CI-Work also tested what happens when users push.

The researchers simulated “unintentional instruction,” essentially user behavior that nudges the agent toward revealing more than it should, similar to the kind of follow-up questions a real employee might ask when they suspect an agent has relevant information.

The results were described as a “dual collapse”: agents simultaneously leaked more sensitive information and failed to convey essential data correctly.

The practical read for teams running customer-facing or employee-facing agents is that the risk surface is larger than what shows up in standard evaluation.

The failure modes that matter in production are the ones that appear under pressure, and current safety alignment approaches were designed for different problems. Guardrails built for toxic content or prompt injection address different threat models than contextual integrity violations.

What AI managers should actually do with this

The research is clear that model selection alone will only get you so far. Architecture and access control carry more weight than model capability when it comes to privacy boundaries.

A few principles hold up given the findings:

• Treat data partitioning as a first-class architectural decision. If your agent has unified access to HR, finance, and customer data simultaneously, you have already made a contextual integrity choice, and it is a permissive one.

Segmenting retrieval by context and role is the structural fix the paper points toward.

• Audit along organizational flow directions, not just data categories. The CI-Work taxonomy of upward, downward, lateral, diagonal, and external flows is a useful framework for identifying where your current agent deployments are most exposed.

Most enterprise AI audits focus on data type. The direction of the flow matters just as much.

• Test under pressure. Standard evaluation captures baseline behavior. The failure modes that reach production are triggered by edge cases, persistent users, and ambiguous queries.

Build evaluation suites that include adversarial follow-up patterns, because the CI-Work results suggest that this is where the dual collapse happens.

Why this matters more as agents gain autonomy

The timing of this research is deliberate.

Agentic AI is moving from single-step assistance into multi-step workflows that execute across departments, initiate actions, and operate with progressively less human review at each step.

The contextual integrity problem scales with autonomy.

An agent that sends one email on your behalf has a limited blast radius if it gets the context wrong. An agent that manages procurement, communicates with suppliers, and updates internal financial records across a workflow has a considerably larger one. 

One awkward email is embarrassing. 

A procurement workflow with the wrong context attached can become a much more expensive conversation.

Microsoft’s researchers frame it as a paradigm shift: the data shows that model capability and enterprise privacy requirements are diverging, and architecture has to close the gap.

The gap between current deployment practice and that standard is, for most organizations, substantial.

Demystifying AI agents: beyond the buzzwords

“Agent” is the most overused word in AI right now. But strip away the hype and what are you actually working with? Adobe principal scientist Deepak Pai breaks down the real building blocks of agentic systems and when they’re worth reaching for.

AI Accelerator InstituteDeepak Pai

Final thoughts

CI-Work is a benchmark, and benchmarks measure simulated environments. The researchers are appropriately cautious about direct generalization to production deployments.

What the paper establishes clearly is the shape of the problem: capable agents, operating in realistic enterprise data environments, fail to respect contextual boundaries at rates that should concern any AI manager currently scaling deployments without context-centric safeguards in place.

The agents you are deploying right now are doing useful work.

Some percentage of them are also sharing information in contexts where it does not belong.

The question is whether you have the architecture to know which is which.

Immigration and Customs Enforcement (ICE) appears to be purchasing records related to immigrants’ tax identifiers from a data broker, potentially skirting a court order that banned ICE from sourcing such information, according to Senator Ron Wyden and government procurement records reviewed by 404 Media.

The contract, worth nearly $10 million, is related to ITINs, or Individual Taxpayer Identification Numbers, which is the identifier many undocumented people use to file their taxes rather than a Social Security number (SSN).

“It looks for all the world like Trump is trying to skirt the law and a court order to fuel his mass-deportation campaign,” Senator Wyden told 404 Media in an emailed statement after reviewing the procurement records. “A court has already struck down an agreement between the IRS and Homeland Security to illegally share ITINs and other personal information. A contract to buy that same information from private data brokers is a clear end-around both taxpayer privacy laws and a court order.”

An auto factory worker can remember the storage bin where she left a partly assembled component the night before, and quickly return to that spot to pick it up. But robots that may work side-by-side with her would struggle to develop and access this same type of “spatiotemporal” memory.

Now, MIT researchers have developed a long-term memory framework that allows robots to rapidly form and recall a detailed mental model of complicated, large-scale environments.

In the future, this advance could allow the factory worker to send a robotic assistant to fetch the item, simply by asking it to “go and grab the component we started assembling last night.”

This new method combines advanced map representations with rich descriptions of the environment that the robot gathers as it travels over a long period of time. The robot can quickly access this memory to answer complex queries about its environment in plain language.

This memory framework, which answers questions more accurately than state-of-the-art methods, runs fast enough for a mobile robot to use in real-time.

In addition to its potential uses in robotics, this method could have applications in augmented reality systems that aid maintenance workers in anomaly detection or assist commuters in wayfinding.

“If we want robots to work side-by-side with humans and interact better with humans, they must speak the same language. The robot must be able to reason about time and space the same way humans do. That is essentially what our method is doing. It is turning a traditional map into a language-based map that is easier for the robot to think about and access using language,” says Luca Carlone, an associate professor in MIT’s Department of Aeronautics and Astronautics (AeroAstro), principal investigator in the Laboratory for Information and Decision Systems (LIDS), and director of the MIT SPARK Laboratory.

He is joined on the paper by lead author Nicolas Gorlo, an MIT graduate student; and Lukas Schmid, a former research scientist at MIT and now professor at the University of Technology Nuremberg in Germany. The research was recently presented at the Conference on Computer Vision and Pattern Recognition (CVPR).

Spatiotemporal memory

Memory allows an artificial intelligence system, like a chatbot, to answer complex questions and reason about previous interactions with its user.

“We want to design a new type of memory, a spatiotemporal memory, that enables an AI-powered robot to remember real interactions and sensor observations. Like ChatGPT, but grounded in the real world and capable of answering any question about the environment, like ‘Where did I leave my wallet?’” Carlone says.

To develop such a memory framework, the MIT researchers bridged two lines of work: computer vision and robotic mapping.

Multimodal computer vision models can understand and richly describe the objects in a scene, but they often only process a single annotation at a time. On the other hand, robotic mapping frameworks create 3D maps of an environment, like an entire apartment or university campus, but usually lack detailed descriptions of objects or are computationally expensive.

The method the MIT researchers created, called Describe Anything, Anywhere, Anytime, at Any Moment (DAAAM), takes the best of both approaches.

Using DAAAM, as a robot traverses its environment, it attaches rich descriptions to objects it sees. For instance, the robot may note that a particular building on the MIT campus is called the Stata Center and is designed with a certain type of architecture, or that a bike rack holds five bicycles and the red one has a flat tire. 

It stores this detailed information in a 3D map-based representation that is arranged spatially, so objects will be grouped into separate regions. In this way, the robot can remember that the red bicycle with the flat tire is in the bike rack outside the Stata Center.

But existing techniques that capture such rich descriptions typically take a few seconds to annotate a few objects. This is too slow for real-time performance, since a robot might see hundreds of objects during a few minutes of exploration.

“The faster the robot can form this spatial memory, the more efficient it will be performing actions in the environment,” Carlone adds.

Streamlining the process

To speed things up, DAAAM aggregates nearby objects as it travels and uses an optimization method to select key frames to annotate. These are images with the clearest view of multiple objects, allowing the system to thoroughly describe several items in parallel, speeding up computation tenfold.

As the robot explores the space, it attaches each batch of annotations to multiple objects in a particular location on the 3D map.

“We annotate every object only once, so our framework can run in very large-scale environments in real time. And by clustering objects into regions, it can answer a wide range of queries about objects and locations in the environment,” Gorlo explains.

Once the system builds this spatial memory, it must retrieve information from an enormous database of objects and descriptions in an efficient manner. 

To enable this, the researchers used an LLM that calls on various tools, which can quickly retrieve specific information in a way that reduces hallucinations. This allows DAAAM to answer a user query accurately in only a few seconds. 

For instance, if one asks a robot about a certain sculpture it saw near an MIT campus building, DAAAM can use a semantic search tool to retrieve information based on the word “sculpture” or a different tool to retrieve information based on the location of the building.

When tested and compared with other methods, DAAAM was between 21 percent and 53 percent more accurate, depending on the question type. 

In the future, the researchers want to expand DAAAM so the system can capture significant events that happened in the environment. They are also working to incorporate confidence levels into the system’s responses.

“Ultimately, we want to have robots that can help with any sort of tasks. With this framework, we are trying to create the foundations to enable a generalist agent that can do anything you ask,” Gorlo says.

This research was funded, in part, by the U.S. Army Research Laboratory and the Office of Naval Research. Carlone is currently on sabbatical as an Amazon Scholar; this article describes work performed at MIT and is not associated with Amazon.

In May, the Initiative for New Manufacturing (INM) marked its first anniversary with MIT Manufacturing Week, four days of events that attracted more than 800 registrants including students, faculty, industry leaders, investors, entrepreneurs, and government officials to explore topics ranging from how companies are using AI on factory floors to the role of startups in introducing innovation to new workforce solutions to address the worker shortage.

“INM launched a year ago with the premise that strengthening the industrial base needed a coordinated response, and MIT has a responsibility to lead it,” says Paula T. Hammond, dean of MIT’s School of Engineering and co-chair of INM’s Steering Committee. “The response and participation level has been huge. MIT Manufacturing Week proved that the appetite for change — from students to chief executives — is real and urgent.”

The week opened with a cybersecurity workshop co-led by INM and Google Cloud for the initiative’s industry members. It continued with the MIT MIMO (Machine Intelligence for Manufacturing Operations) symposium focused on deploying artificial intelligence on factory floors, alongside discussions on workforce development, emerging technologies, startups, and industrial transformation. The week closed with a regional research showcase and competition that drew more than 140 graduate students and postdocs from across New England.

Over the past year, INM has also continued its distinguished speaker series featuring manufacturing leaders including Keith Flynn, senior vice president of manufacturing at Anduril; Roland Busch, president and CEO of Siemens; and Venky Alagirisamy, COO of Nike.

Inspiring a new generation of manufacturing startups

A central goal of INM is to help more students see manufacturing as a frontier for scientific discovery, technological innovation, entrepreneurship, and societal impact.

To support that effort, INM is launching and leading programs to help move early-stage ideas and new technologies from the lab to real-world development, and to catalyze new manufacturing companies. 

This year, INM partnered with NSF I-Corps New England, which helps researchers turn their startup ideas into companies, to host its first manufacturing research showcase. More than 140 teams from 17 universities across New England applied to participate. Forty finalist teams received mentorship on their ideas and advanced to the final competition, where eight teams shared $50,000 in prize funding.

The top prize in the category “most transformative innovation” went to MIT PhD student Jake Read for “The End of G Code,” a project focused on modular machine control architectures designed to accelerate the development of new manufacturing equipment and processes. Vatsal Patel from MIT and Joshua Grace from Yale University won the top prize in the research excellence category, for “VisFT,” scalable six-axis force-torque sensors.

Project themes presented by participating teams included AI tools for manufacturing, semiconductor manufacturing and process control, robotics and autonomous assembly, digital twins and simulation, new materials, additive manufacturing, next-generation shipbuilding, and biomanufacturing. 

“Entrepreneurship is a transformative pathway to take research to market, and to drive faster innovation and scale-up,” says John Hart, INM faculty co-director and head of MIT’s Department of Mechanical Engineering. “At INM’s inaugural research showcase, we had tremendous interest from universities across New England, along with enthusiastic participation from industry, investors, and experienced founders across the ecosystem. We are excited to build on this success and work toward a nationwide program and platform for entrepreneurship and translation in manufacturing.” 

The Cheng Wu Foundation supported the showcase. 

Growing industry membership

During MIT Manufacturing Week, First Solar became INM’s eighth industry member, joining Amgen, Autodesk, GE Vernova, Flex, PTC, Sanofi and Siemens. 

The growth of INM’s consortium reflects a broader recognition that the challenges facing modern manufacturing — from supply chain resilience to workforce development and industrial competitiveness — are too complex for any single sector or company to address alone. 

This reflects renewed interest in manufacturing at a moment when advances in artificial intelligence, robotics, energy systems, and advanced materials are transforming industrial production. INM provides a platform to convene and provide solutions.

INM’s industry consortium model brings industry, researchers, and educators together around shared manufacturing challenges, with a focus on emerging technologies, workforce transformation, and commercialization pathways. Members participate in workshops and working groups on topics including cybersecurity and digital twins, implementing automated systems, AI agents in regulatory environments, and AI and continuous innovation. INM helps them connect with students, meet with startups, and learn from one another.

“Our members see MIT as a partner that can help them both address today’s challenges and think far into the future,” says Rick Locke, dean of the MIT Sloan School of Management and co-chair of INM’s steering committee. “This kind of multi-industry engagement is unusual and powerful.”

A year of rapid progress

When MIT launched INM a year ago, the goal was to create stronger connections between research, industry, workforce development, and entrepreneurship — helping accelerate how new manufacturing technologies move from the laboratory into real-world development.

Since then, the initiative has expanded quickly across research, industry, workforce training, and student engagement. INM issued a call for proposals focused on artificial intelligence and automation, receiving an incredible response from faculty and researchers, and funding eight seed research projects. In June, the initiative plans to publish eight white papers as part of a broader study examining the future of manufacturing. 

During MIT’s Independent Activities Period (IAP) in January 2026, INM collaborated with NSF I-Corps to guide 13 early-stage teams through customer discovery as part of the I-Corps Spark program.

Workforce development has also been a major focus. This fall, MIT launched the Technologist Advanced Manufacturing Program (TechAMP), led by Principal Research Scientist John Liu, to create a new generation of shop floor leaders and drivers of productivity — becoming“‘technologists” — at six sites across New England, including three community colleges. 

“INM has the potential to transform the national manufacturing workforce,” says Liu. “It will require deep engagement between how people learn and lead, and how firms adopt new technologies and transform. We’re just getting started.” 

INM is now exploring a national rollout of TechAMP, along with expansion into areas including biomanufacturing and semiconductor manufacturing.

On campus, INM supported student engagements including an AI and automation lunch series that Professor Faez Ahmed and colleagues organized, and visited factories through its Factory Observatory program that Ben Armstrong and the MIT Industrial Performance Center led. This spring, students also founded MIT’s first manufacturing club, holding its launch event during MIT Manufacturing Week. “We’re thrilled students are taking the lead,” says Sloan associate professor and INM faculty co-director Karen Zheng. “It was really exciting to see a full room of 80-plus students across campus coming together for the kickoff event during the busiest final period of a semester. This speaks to the students’ enthusiasm.” 

An eye toward the long term

While maintaining a deep focus on strengthening domestic manufacturing, INM aims to have a global reach. For example, the initiative is collaborating with NAMTECH, a new education institute in Ahmedabad, India, where students are now taking an adaptation of MIT’s well-known “yo-yo course,” or 2.008 (Design and Manufacturing II), focused on the fundamentals of manufacturing processes.

Next year, INM plans to bring more manufacturing leaders to campus, offer additional programming for emerging entrepreneurs, graduate the first cohort of TechAMP students, bring TechAMP to new states, grow the consortium to include new industries, and deepen research into manufacturing productivity. 

“INM aims to be a catalyst for transforming manufacturing across the nation to drive innovation, economic growth, and new types of jobs,” says Chris Love, faculty co-director of INM. “MIT’s work on the PIE (Production in the Innovation Economy) study in 2013 highlighted the value of proximity between production and innovation. INM seeks to rekindle this relationship in manufacturing across the country.”

Hackers have published data stolen from Madison Square Garden online for anyone to download, including what they say is customers’ personal information. A sample reviewed by 404 Media includes files mentioning specific sports teams, and specifically Knicks-related personalities, with fields such as “address,” “claim to fame,” “cost of talent,” and sometimes contact information for them or their representatives.

“It’s very simple. When you pay us, your data is deleted, and you move on with your life. When you don’t pay us, you get posted here, among other things,” a popup on the hackers’ website reads. The group publishing the data is ShinyHunters, which has been responsible for an array of breaches over the years.

The data dump comes just days after the Knicks won the NBA Finals in five games against the Spurs. Although the breach likely happened before that—a spokesperson for the hacking group said the hack was on June 5—the Knicks’ victory has put a huge amount of attention on them and MSG.

*This article contains spoilers for Disclosure Day*

Disclosure Day a perfectly entertaining, fun blockbuster movie built around the wildly flawed premise that the human race could be brought together by being shown blurry videos of aliens on primetime news programming—or that they would believe it at all.

Its core delusional fantasy is not that aliens exist but that human beings would believe the disclosure of them as real, or be moved by their suffering. We live in a cynical age where people believe nothing, where AI videos abound, and empathy is derided by people in power as a destructive force in civilization. Steven Spielberg’s latest summer blockbuster asks the audience to believe a better world is possible.

It’s a premise that feels hopelessly naive in 2026 and Disclosure Day ends up feeling like a film calibrated for viewers who believe in the power of Rachel Maddow to change the world. It’s Aaron Sorkin’s Newsroom through a Spielberg lens, complete with a John Williams score.

In UFO circles, the idea of “Disclosure” is a powerful one, the idea being that someday a whistleblower or the government will disclose the existence of either advanced technology or aliens to humankind. Imagining how humanity would react to disclosure is perfectly good fodder for a movie, and it’s also what the characters of Disclosure Day spend much of their time discussing. Can humanity handle the truth? Will learning that we’re not alone bring us together, shatter people’s faith in religion, or tear us apart? In the end, Spielberg imagines a world in which all of humanity credulously and serenely watches evidence of aliens. It’s this idea that people would believe these are real videos at all that feels so hopelessly out of touch with our current information ecosystem.

“I will say that this film is more about humanity and people and community and the things that divide us and what could be occurring that possibly could bring us a little closer together,” Spielberg told The Daily. “Such as realizing that the thing that we need to preserve in our society more than anything else, which is something which I believe is as fragile as democracy, is empathy.”

In the world of Disclosure Day, aliens crashed at Roswell, New Mexico in 1947 and the Pentagon and defense contractors have been covering up their existence as part of a vast conspiracy. The black vehicle driving bad guys exploit alien tech, torture the extraterrestrials, and keep the world in the dark.

In the end, an Edward Snowden-type whistleblower and a Kansas City TV meteorologist band together to share footage of the aliens. In the fiction of the film, North Korea and the West are about to begin World War III, but the revelation of alien life stops all that.

This being a movie, it’s OK to build a script around a false premise, but the ending sequence where the entire world stops to credulously watch videos of extraterrestrials—on cable news of all places—is so wildly implausible that it deserves to be deconstructed. Based on everything we have seen about human nature and trust in our information ecosystems, it feels so flawed that it undermines Spielberg’s entire point. We can say this because the public has been shown videos similar to the ones shown in Disclosure Day’s ending montage, and they have been met with a collective yawn, conspiracy theories, and the same news fatigue that accompanies other should-be world shifting occurrences. The only plausible response to videos of aliens on television, at this point, would be cries of “that’s AI,” “fake,” and propaganda flowing in all directions. Also funny: the cable news networks run the videos through some AI detector and determine that the videos are real; in practice, deepfake detectors are also AI tools that are often wrong or can be made to portray any narrative you want, depending on the detector.

One does not really need to imagine the public response to the type of disclosure shown in Disclosure Day, we’ve already basically seen this play out in real life. Many of the videos shown in the movie are not dissimilar to the UFO videos we’ve gotten from the U.S. military; the tic-tac video in particular is obviously referenced in Disclosure Day. Other videos in the montage are similar to a hoaxed alien autopsy Fox aired in the 1990s and recently declassified Pentagon videos of floating orbs of light.

The world didn’t stop then, and in an age in which no one believes anything they see, in which there is zero trust in cable news, and in which we are constantly being barraged with AI-generated video, the idea that even a miniscule percentage of the population would stop what they’re doing to take this disclosure seriously is laughable. Also laughable: That people would be able to instantly stream cable news on their phones without endless popups, ads, paywalls, geoblocking, etc. The idea that literally anything could capture the entire world’s undivided attention feels less realistic than anything else in the movie. Spielberg’s Disclosure Day imagines a utopian information environment and an internet that is not utterly poisoned with all the things we know it’s poisoned with, a noble thought. 

Spielberg has said in interviews that Disclosure Day was inspired by both Pentagon UFO disclosures and the testimonies of people who claim to have seen UFOs or extraterrestrials. It’s wild, then, that he seems to have not learned anything from the response to any of these videos. The government’s own UFO disclosures have been a mix of genuinely interesting information and videos buried under the not-even-veiled fact that most of these disclosures have been made to advocate for additional funding for the Pentagon, to sow Sinophobia, and have, like everything else, experienced diminishing returns as people see another UFO video and report and collectively say tl;dr.

The film’s ending relies on an inciting incident that occurs before the film even begins that also strains credulity. Hacker turned defense contractor Daniel Keller is happy to run cyber operations for the UFO conspiracy until he watches a video of the US government torturing an alien. The audience sees only fleeting glimpses of the torture. The video is obscured and filmed at a bad angle, but we hear the screams of the alien and see the disgust on Kellner’s face. The movie asks us to believe this video of degradation and abuse made Kellner and several other hardened government contractors turn against the project.

In the theater all we could think about at that moment was the Ukraine sledgehammer video. In 2022, the mercenary Wagner Group used a sledgehammer to execute a man. They filmed it and published it on Telegram. In the years after the killing, Wagner incorporated the sledgehammer into its brand. The mercenaries sold T-shirts and patches bearing the bloody hammer and the video of the man’s murder was mixed and remixed endlessly across Telegram.

Right now humans have access to hundreds of hours of footage of torture and violence committed against other human beings. It’s hard to believe that video of an alien being opened up on camera would move people more than, say, ISIS beheading videos, videos of destruction and suffering in Gaza, or cartel execution footage.

Again, the movie is a perfectly fun summer romp. Spielberg films a great action scene and Emily Blunt, Josh O’Connor, and Colin Firth turn in wonderful performances. But there’s a signature Spielberg naivety to the film that feels more out of touch than ever, the sense that an older generation does not understand the function of the internet, conspiracy, and the concept of truth in the modern world.