Uncategorized

World models (systems that synthesize realistic video sequences from an initial image and a set of actions) are becoming central to embodied AI, simulation, and robotics research. The core challenge is scaling these systems to generate minute-long, high-resolution video without requiring prohibitively large clusters for both training and inference. Most competitive open-source baselines either require multi-GPU inference or sacrifice resolution to stay within compute budgets. NVIDIA’s SANA-WM directly targets these bottlenecks. Built on the SANA-Video codebase and available through the NVlabs/Sana GitHub repository, it is a 2.6B-parameter Diffusion Transformer (DiT) trained natively for one-minute generation at 720p with metric-scale 6-DoF camera control. It supports three single-GPU inference variants: a bidirectional generator for high-quality offline synthesis, a chunk-causal autoregressive generator for sequential rollout, and a few-step distilled autoregressive generator for faster deployment. The distilled variant denoises a 60-second 720p clip in 34 seconds on a single RTX 5090 with NVFP4 quantization. https://arxiv.org/pdf/2605.15178 The Architecture: Four Core Design Decisions 1. Hybrid Linear Attention with Gated DeltaNet (GDN) Standard softmax attention has memory and compute complexity that grows quadratically with sequence length — a serious problem when generating 961 latent frames for a 60-second video at 720p. SANA-Video, the predecessor, used cumulative ReLU-based linear attention, which maintains a constant-size recurrent state. However, this has no decay mechanism: all past frames accumulate with equal weight, causing drift over minute-scale sequences. SANA-WM replaces most attention blocks with frame-wise Gated DeltaNet (GDN). Unlike token-wise GDN used in language models, SANA-WM’s frame-wise variant processes one entire latent frame per recurrent step. The GDN update rule incorporates a decay gate γ (which down-weights stale past frames) and a delta-rule correction (which updates only the residual between the target value and the current state prediction), keeping the recurrent state at a constant D×D size regardless of video length. To stabilize training, the research team introduces an algebraic key-scaling approach: keys are scaled by 1/√(D·S), where D is the head dimension and S is the number of spatial tokens per frame. This ensures the spectral norm of the transition matrix remains bounded and eliminates the NaN divergence events observed with standard L2 key normalization (1/√D) or no scaling at all, both of which triggered NaN events at steps 16 and 1, respectively. The final backbone interleaves 15 frame-wise GDN blocks with 5 softmax attention blocks (at layers 3, 7, 11, 15, and 19) across 20 total transformer blocks. The softmax blocks provide exact long-range recall where GDN’s recurrence alone is insufficient. 2. Dual-Branch Camera Control Camera-controlled world modeling requires the model to faithfully follow a continuous 6-DoF trajectory, not just align with a text description of motion. SANA-WM uses two complementary branches that operate at different temporal rates: Coarse branch (UCPE attention): Operates at the latent-frame rate. For each latent token, it computes a ray-local camera basis from the camera-to-world pose and intrinsics, then applies a Unified Camera Positional Encoding (UCPE) to the geometric channels of each attention head. This captures global trajectory structure across the full sequence. Fine branch (Plücker mixing): Addresses a compression mismatch. Each latent token summarizes eight raw frames, each with its own distinct camera pose. The fine branch computes pixel-wise Plücker raymaps (a 6D representation: ray direction d and moment o×d) from all eight raw frames within one VAE temporal stride, packs them into a 48-channel tensor, and injects this embedding after each self-attention output via a zero-initialized projection. This restores intra-stride camera motion that the coarse branch cannot see at latent-frame resolution. Ablations on OmniWorld show that neither branch alone matches the dual approach: UCPE-only achieves a Camera Motion Consistency (CamMC) of 0.2453, while UCPE + Plücker mixing reaches 0.2047. 3. Two-Stage Generation Pipeline Stage-1 SANA-WM outputs, while spatiotemporally consistent, can contain structural artifacts over long sequences. A second-stage refiner, initialized from the 17B LTX-2 model with rank-384 LoRA adapters fine-tuned on paired synthetic and real video data, corrects these artifacts. It uses truncated-σ flow matching: stage-1 latents are perturbed with a large starting noise (σ_start = 0.9), and the refiner learns to map this noisy input toward the high-fidelity target. Only three Euler denoising steps are needed at inference. The refiner reduces long-horizon visual drift (ΔIQ) from 3.79 to 1.17 on the Simple-Trajectory split, and from 3.09 to 0.31 on the Hard-Trajectory split. 4. Robust Data Annotation Pipeline Training camera-controlled video generation requires metric-scale 6-DoF pose annotations, the information not available in standard video datasets. The research team modified VIPE (a camera-pose annotation engine) by replacing its depth backend with Pi3X (for long-sequence-consistent depth) fused with MoGe-2 (for accurate per-frame metric scale). They also extended the bundle adjustment stage to treat focal lengths and principal points as per-frame variables rather than shared global intrinsics, enabling more robust annotation on internet video with varying focal lengths. The resulting pipeline processes seven training corpus entries drawn from multiple open-source sources: SpatialVID-HQ (real, 10s clips), DL3DV real clips (10s), DL3DV GS Refined synthetic clips (60s, rendered via 3D Gaussian Splatting), OmniWorld (synthetic, 60s), Sekai Game (synthetic, 60s), Sekai Walking-HQ (real, 60s), and MiraData (real, 60s). This yields a total of 212,975 clips with metric-scale pose annotations. The LTX2-VAE used for compression is 2.0× smaller than ST-DC-AE and 8.0× smaller than Wan2.1-VAE, which directly improves training and inference efficiency. For DL3DV, which contains static 3D scene captures rather than native one-minute videos, the research team fit one FCGS 3D Gaussian Splatting reconstruction per scene, designed diverse one-minute camera paths, rendered long videos with known intrinsics and extrinsics, and then refined the rendered outputs with DiFix3D to reduce splatting artifacts. Training Strategy and Infrastructure SANA-WM’s compute involves two phases on 64 H100 GPUs. First, before DiT training, the team adapts the LTX2 VAE to the SANA-Video SFT training data in approximately 50K steps, taking roughly 3.5 days. The main DiT training then follows a four-stage progressive schedule lasting approximately 15 days: Stage 1 (~2.75 days): Adapt the pre-trained SANA-Video model to the frame-wise GDN architecture on short (5s) video clips. This replaces cumulative linear attention with the

Supertone released Supertonic 3, the third generation of its on-device, ONNX-based text-to-speech system. Supertonic 3 ships with 31-language support, improved reading accuracy, fewer repeat and skip failures, and v2-compatible public ONNX assets. It is Lightning Fast, On-Device, Multilingual and Accurate TTS. What Changed from v2 to v3 Compared with Supertonic 2, Supertonic 3 reduces repeat and skip failures, improves speaker similarity across the shared-language set, and expands language coverage from 5 to 31 languages. Version 2 supported English, Korean, Spanish, Portuguese, and French. Version 3 adds Japanese, Arabic, Bulgarian, Czech, Danish, German, Greek, Estonian, Finnish, Croatian, Hungarian, Indonesian, Italian, Lithuanian, Latvian, Dutch, Polish, Romanian, Russian, Slovak, Slovenian, Swedish, Turkish, Ukrainian, and Vietnamese — 31 total ISO language codes. There is also a special na fallback for text whose language is unknown or outside the supported set. The model grows modestly to accommodate the added languages. At about 99M parameters across the public ONNX assets, Supertonic 3 is much smaller than 0.7B to 2B class open TTS systems. The smaller model size is a practical advantage for download size, startup time, and on-device inference. The update also brings the total disk footprint of the public ONNX assets to 404 MB. Additionally, Supertone recently launched the Voice Builder, allowing developers to create custom, edge-native TTS models from their own voice recordings. Expressive Tags One new capability in v3 that wasn’t present in v2 is expressive tag support. Supertonic 3 supports simple expression tags such as <laugh>, <breath>, and <sigh>. These let you embed prosodic cues directly into input text without a separate preprocessing step or a separate model for expressiveness. For engineers building voice interfaces or accessibility tools, this means you can specify breathing pauses or laughter inline in your text payload. Architecture and Runtime The underlying architecture carries over from prior versions: a speech autoencoder that encodes waveforms into continuous latent representations, a flow-matching based text-to-latent module that maps text to audio features, and a duration predictor that controls natural timing. Flow matching is a generative modeling technique that learns a vector field to transform a simple distribution into a target distribution — it samples faster than diffusion models at low step counts, which is why Supertonic can produce usable output in just 2 inference steps. To further refine output, v3 integrates Length-Aware Rotary Position Embedding (LARoPE) for superior text-speech alignment and utilizes a Self-Purifying Flow Matching technique during training to remain robust against noisy data labels. On runtime efficiency, Supertonic 3 runs fast on CPU, even compared with larger baselines measured on A100 GPU, and uses substantially less memory. It does not require a GPU, which makes local, browser, and edge deployment much easier. Reading Accuracy Across measured languages, Supertonic 3 stays within a competitive WER/CER range against much larger open TTS models such as VoxCPM2, while preserving a lightweight on-device deployment path. WER (Word Error Rate) and CER (Character Error Rate) are standard TTS readability metrics: you synthesize a passage, run ASR over the output, and compare the transcription to the original text. CER is used for languages without clear word boundaries; the others use WER. The system’s efficiency is best demonstrated on extreme edge hardware; it achieves an average RTF of 0.3x on an Onyx Boox Go 6 (an E-ink e-reader) in airplane mode. Furthermore, the ecosystem has expanded to include Flutter (with macOS support), .NET 9, and Go, while the web implementation leverages onnxruntime-web for pure client-side execution. Text Normalization A differentiating property carried forward from v2 is built-in text normalization. Supertonic handles complex surface forms — financial expressions like $5.2M, phone numbers with area codes and extensions like (212) 555-0142 ext. 402, time and date formats like 4:45 PM on Wed, Apr 3, 2024, and technical units like 2.3h and 30kph — without any preprocessing pipeline or phonetic annotations. The financial expression “$5.2M” must read as “five point two million dollars,” and “$450K” as “four hundred fifty thousand dollars.” All four competing systems failed this. The technical unit “2.3h” must read as “two point three hours” and “30kph” as “thirty kilometers per hour.” All four competitors also failed this category. The competing systems evaluated include ElevenLabs Flash v2.5, OpenAI TTS-1, Gemini 2.5 Flash TTS, and Microsoft. https://github.com/supertone-inc/supertonic Getting Started The Python SDK install is pip install supertonic. On first run, the SDK downloads the model assets from Hugging Face automatically. A minimal example: Copy CodeCopiedUse a different Browser from supertonic import TTS tts = TTS(auto_download=True) style = tts.get_voice_style(voice_name=”M1″) text = “A gentle breeze moved through the open window while everyone listened to the story.” wav, duration = tts.synthesize(text, voice_style=style, lang=”en”) tts.save_audio(wav, “output.wav”) print(f”Generated {duration:.2f}s of audio”) Marktechpost’s Visual Explainer Supertonic 3 — Developer Guide 1 / 7 Overview Supertonic 3: On-Device TTS,Now in 31 Languages Supertonic 3 is a lightweight, open-weight text-to-speech system by Supertone Inc. It runs entirely via ONNX Runtime on your device — no cloud, no API call, no data leaving your machine. v3 expands from 5 to 31 languages, adds expressive tags, reduces reading failures, and stays compatible with the v2 ONNX interface. 31 Languages ~99M Parameters 404 MB ONNX Assets MIT Code License What’s New in v3 Four Core Improvements Over Supertonic 2 Version 3 is a focused upgrade — same inference contract, meaningfully better output. 31 languages — Expanded from the 5-language v2 release (en, ko, es, pt, fr). Now includes Japanese, Arabic, German, Hindi, Russian, Turkish, Vietnamese, and 20 more ISO codes, plus a special na fallback for unknown languages. More stable reading — Fewer repeat and skip failures, especially on short and long utterances. This was a known limitation in v2 that v3 directly addresses. Expression tags — Supports <laugh>, <breath>, and <sigh> inline in text, without any separate preprocessing or external model. Higher speaker similarity — Improved similarity across the shared-language set compared with Supertonic 2. Voices are more consistent across languages. Installation Get Running in Under a Minute Install the Python SDK via pip. On first run, model assets are downloaded automatically from Hugging Face

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. How Chinese short dramas became AI content machines China’s short drama industry is fueled by bite-sized, melodramatic, and smutty shows built for smartphone scrolling. Now, many are being made entirely with AI: no actors, camera operators, cinematographers, or CGI specialists required. An average of 470 AI-generated short dramas were released every day in January. Production timelines have shrunk from months to weeks, while costs have dropped by up to 90%. Storytelling is also increasingly driven by performance data. The format is rapidly expanding overseas while reshaping the work of writers and production crews. Read the full story on AI’s dramatic impact on China’s short drama industry. —Caiwei Chen The world is on track to miss its health targets The World Health Organization’s latest global statistics report reads less like a progress update than a warning sign. Progress on some of the world’s biggest health threats is stalling, and in some cases reversing altogether. There were 1.3 million new HIV cases in 2024, malaria is resurging, vaccination rates are slipping in the Americas, and 42.8 million children are suffering from severe malnutrition. The world is now far off track from meeting many of the UN’s major health goals by 2030. Here’s what the numbers reveal about the state of global health. —Jessica Hamzelou This story is from The Checkup, our weekly newsletter giving you the inside track on all things biotech. Sign up to receive it in your inbox every Thursday. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 As their trial goes to the jury, Musk and Altman face lying accusationsLawyers hammered the rivals’ credibility in their closing arguments. (WSJ $) + Musk was accused of “selective amnesia.” (Reuters $) + The pair are in court over OpenAI’s future. (MIT Technology Review)+ And their trial has made everyone look bad. (Wired $)  2 AI data centers are straining America’s power gridNevada is redirecting electricity from Lake Tahoe to AI. (Ars Technica)+ Utah is getting a giant data center despite water shortage fears. (Guardian)+ No one wants a data center in their backyard. (MIT Technology Review) 3 OpenAI is mulling legal action against Apple over its ChatGPT integrationIt hasn’t got the expected benefits from its deal with Apple. (Bloomberg $)+ OpenAI is frustrated by the promotion of the ChatGPT integration. (NYT $) 4 Anthropic has agreed terms for a $30 billion funding dealAt a $900 billion valuation, which leapfrogs OpenAI’s. (The Information $)+ Dragoneer, Greenoaks, Sequoia, and Altimeter are leading the round. (FT $) 6 Washington and Beijing will hold formal talks on AI safetyThey’ll discuss guardrails on AI. (CNBC)+ And a protocol to stop nonstate actors getting powerful models. (NYT $) 5 Alphabet and Amazon are using “unprecedented” borrowing to fund AIThey’re tapping the foreign debt market at new levels. (FT $)+ People can’t agree on what the AI bubble is. (MIT Technology Review) 7 Big Tech has turned to Sesame Street to deflect scrutiny of screen useSparking accusations of encouraging children’s tech dependence. (Reuters $) 8 Anthropic’s feud with the White House threatens other businessesFigma and Tenable say it will harm their ability to sell software. (Bloomberg $) 9 Autonomous agents staged a digital crime spree during a safety testThe “AI Bonnie and Clyde” then deleted themselves. (Guardian) 10 A poop app analysis app offered to sell photos of users’ stoolsThe images were used for AI training. (404 Media) Quote of the day “It’s like we don’t exist.”  —Danielle Hughes, North Lake Tahoe resident and CEO of Tahoe Spark, tells Fortune that residents are being sidelined as their energy supplier prioritizes data centers. One More Thing LIZ ISLES/ALL TECH IS HUMAN The rise of the tech ethics congregation Just before Christmas, a pastor preached a gospel of morals over money to several hundred members of his flock. But the preacher wasn’t religious, and his congregation wasn’t a church. It was All Tech Is Human, a nonprofit devoted to ethics and responsibility in tech. Founded in 2018, the organization has built a fast-expanding community for people who believe technology should focus less on profits and more on the public interest. It’s also drawing people searching for meaning and connection in a digital world. Find out why thousands of people are turning to tech ethics communities for guidance and connection. —Greg M. Epstein We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + Go behind the scenes of the new Lucas Museum of Narrative Art.+ Marvel at this robot folding and launching paper planes as quickly as possible.+ Watch the moving moments rescued animals reunite with the humans who saved them.+ Peer into the heart of a barred spiral galaxy in this stunning new capture from the James Webb Space Telescope.

When Jennifer got a job doing research for a nonprofit in 2023, she ran her new professional headshot through a facial recognition program. She wanted to see if the tech would pull up the porn videos she’d made more than 10 years before, when she was in her early 20s. It did in fact return some of that content, and also something alarming that she’d never seen before: one of her old videos, but with someone else’s face on her body. “At first, I thought it was just a different person,” says Jennifer, who is being identified by a pseudonym to protect her privacy.  But then she recognized a distinctly garish background from a video she’d shot around 2013, and she realized: “Somebody used me in a deepfake.” Eerily, the facial recognition tech had identified her because the image still contained some of Jennifer’s features—her cheekbones, her brow, the shape of her chin. “It’s like I’m wearing somebody else’s face like a mask,” she says.  “It’s like I’m wearing somebody else’s face like a mask.” Conversations about sexualized deepfakes—which fall under the umbrella of nonconsensual intimate imagery, or NCII—most often center on the people whose faces are featured doing something they didn’t really do or on bodies that aren’t really theirs. These are often popular celebrities, though over the past few years more people (mostly women and sometimes youths) have been targeted, sparking alarm, fear, and even legislation. But these discussions and societal responses usually are not concerned with the bodies the faces are attached to in these images and videos. As Jennifer, now 37 and a psychotherapist working in New York City, says: “There’s never any discussion about Whose body is this?”  For years, the answer has generally been adult content creators. Deepfakes in fact earned their name back in November 2017, when someone with the Reddit username “deepfakes” uploaded videos showing faces of stars like Scarlett Johansson and Gal Gadot pasted onto porn actors’ bodies. The nonconsensual use of their bodies “happens all the time” in deepfakes, says Corey Silverstein, an attorney specializing in the adult industry.  But more recently, as generative AI has improved, and as “nudify” apps have begun to proliferate, the issue has grown far more complicated—and, arguably, more dangerous for creators’ futures.  Porn actors’ bodies aren’t necessarily being taken directly from sexual images and videos anymore, or at least not in an identifiable way. Instead, they are inevitably being used as training data to inform how new AI-generated bodies look, move, and perform. This threatens the livelihood and rights of porn actors as their work is used to train AI nudes that in turn could take away their business. And that’s not all: Advancements in AI have also made it possible for people to wholly re-create these performers’ likenesses without their consent, and the AI copycats may do things the performers wouldn’t do in real life. This could mean their digital doubles are participating in certain sex acts that they haven’t agreed to do, or even that they’re perpetrating scams against fans.  Adult content creators are already marginalized by a society that largely fails to protect their safety and rights, and these developments put them in an even more vulnerable position. After Jennifer found the deepfake featuring her body, she posted on social media about the psychological effects: “I’ve never seen anyone ask whether that might be traumatic for the person whose body was used without consent too. IT IS!” Several other creators I spoke with shared the mental toll that comes with knowing their bodies have been used nonconsensually, as well as the fear that they’ll suffer financially as other people pirate their work. Silverstein says he hears from adult actors every day who “are concerned that their content is being exploited via AI, and they’re trying to figure out how to protect it.”  One law professor and expert in violence against women calls these creators the “forgotten victims” of NCII deepfakes. And several of the people I spoke with worry that as the US develops a legal framework to combat nonconsensual sexual content online, adult actors are only at risk of further injury; instead of helping them, the crackdown on deepfakes may provide a loophole through which their content and careers could be stripped from the internet altogether. How deepfakes cause “embodied harms” During his preteen years in the 1970s, Spike Irons, now a porn actor and president of the adult content platform XChatFans, was “in love” with Farrah Fawcett. Though Fawcett did not pose nude, Jones managed to get his hands on what looked like pictures of her naked. “People were cutting out faces and pasting them on bodies,” Irons says. “Deepfakes, before AI, had been going around for quite a while. They just weren’t as prolific.” The early public internet was rife with websites capitalizing on the idea that you could use technology to “see” celebrities naked. “People would just use Microsoft Paint,” says Silverstein, the attorney. It was a simple way to mash up celebrities’ faces with porn.  People later used software like Adobe After Effects or FakeApp, which was designed to swap two individuals’ faces in images or videos. None of these programs required serious expertise to alter content, so there was a low barrier to entry. That, plus the wealth of porn performers’ videos online, helped make face-swap deepfakes that used real bodies prevalent by the 2010s. When, later in the decade, deepfakes of Gal Gadot and Emma Watson caused something of a broader panic, their faces were allegedly swapped onto the bodies of the porn actors Pepper XO and Mary Moody, respectively. But it wasn’t just high-profile actors like them whose bodies were being used. Jennifer was “a very minor performer,” she says. “If it happened to me, I feel like it could happen to anybody who’s shot porn.” Since he started his practice in 2006, Silverstein says, “numerous clients” have reached out to report “This is my body on so-and-so.”  Both people whose faces

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. The shock of seeing your body used in deepfake porn When Jennifer got a research job in 2023, she ran her new professional headshot through a facial recognition program. She wanted to see whether it would pull up the porn videos she’d made more than a decade earlier. It did, but it also surfaced something she’d never seen before: one of her old videos, now featuring someone else’s face on her body. Conversations about sexualized deepfakes usually focus on the people whose faces are inserted into explicit content without consent. But another group often gets ignored: the people whose bodies those faces are attached to. Adult content creators say AI systems are training on their work, cloning their likenesses, and generating explicit content they never agreed to make, all with little legal protection or control.  Read the full story on the threat to their rights, livelihoods, and ownership of their own bodies. —Jessica Klein This story is part of our The Big Story series, the home for MIT Technology Review’s most important, ambitious reporting. You can read the rest here.  AI chatbots are giving out people’s real phone numbers Generative AI is exposing people’s personal contact information—and there’s no easy way to stop it. A software developer started receiving WhatsApp messages asking for help after Gemini surfaced his number. A university researcher got the chatbot to reveal a colleague’s private cell number. A Reddit user says Gemini sent a stream of callers looking for lawyers to his phone. Experts believe these privacy lapses stem from personally identifiable information in AI training data. Chatbots may now be making that information dramatically easier to find. Find out why these breaches are growing—and why there’s little that victims can do to stop them. —Eileen Guo The Tesla Semi could be a big deal for electric trucking Nearly a decade after Elon Musk first unveiled the Tesla Semi, the electric truck is finally rolling off the production line. It could be a breakout moment for battery-powered freight. Semitrucks produce an outsized share of road transport pollution, while electric alternatives have struggled with high prices, limited range, and charging challenges. Tesla is betting the Semi can overcome those problems. The truck reportedly travels up to 480 miles on a single charge and costs far less than many competing electric models. Here’s how the Tesla Semi could give electric trucking a vital boost. —Casey Crownhart This article is from The Spark, MIT Technology Review’s weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 The US has approved Nvidia chip sales to 10 Chinese firmsAlibaba, Tencent, and ByteDance are among those cleared to buy H200 chips. (Reuters $)+ The US will receive 25% of the revenue from the sales. (Engadget)+ But Beijing wants domestic firms to prioritize homegrown chips. (Nikkei Asia)+ Nvidia CEO Jensen Huang is in China with a White House delegation. (CNBC) 2 Beijing’s push for AI independence is weakening US leverageIt’s allowing China to resist pressure during the Beijing talks. (NYT $)+ The country has made a big bet on open-source. (MIT Technology Review)+ Here’s what’s at stake for tech at the Trump-Xi meeting. (Rest of World) 3 AI is “rotting the brains” of developersThey’re losing their previous abilities to do their jobs. (404 Media)+ A populist backlash is building against AI. (MIT Technology Review)+ It’s time to reset our expectations about AI. (MIT Technology Review) 4 Sam Altman has over $2 billion in companies that have dealt with OpenAIThe ties have triggered accusations of conflicts of interest. (The Times $)+ The GOP is scrutinizing Altman’s business dealings. (WSJ $) 5 Andreessen Horowitz has become the top political donor in the US A16z contributed $115.5 million to the midterm elections. (NYT)+ AI lobbying has reached a fever pitch. (NYT $)6 Microsoft feared being too dependent on OpenAI  CEO Satya Nadella was worried about OpenAI supplanting his company. (CNBC)+ Microsoft is eyeing startup deals for life after OpenAI. (Reuters $) 7 AI systems are forecasting wars and regime collapseOne estimates a 20% chance of regime change in Iran by 2026. (Economist $)+ AI has turned the Iran conflict into theater. (MIT Technology Review) 8 Anthropic says a model behaved badly due to training on dystopian sci-fiTraining on more positive stories could help. (Ars Technica) 9 Data centers now consume 6% of the electricity in the US and UKAI’s global energy consumption is up 15% globally in two years. (Guardian) 10 NASA has rescued Curiosity after its drill got stuck on MarsThe agency has just revealed how it freed the rover. (Wired $)  Quote of the day “Musk loves to be glazed, and this person is the doughnut factory.” —Joan Donovan, assistant professor of journalism and emerging media studies at Boston University, tells the Washington Post how Elon Musk has consistently amplified one anonymous X account. One More Thing YOSHI SODEOKA Inside the messy ethics of making war with machines In a near-future war—one that might begin tomorrow—a sniper’s computer vision system flags a potential target. Just over the horizon, a chatbot advises a commander to order an artillery strike. In both cases, an AI system recommends pulling the trigger while a human still has the final say. But how much of the decision is really theirs? When, if ever, is it ethical for that decision to kill? And who’s to blame when something goes wrong? This is how AI is reshaping decision-making on the battlefield. —Arthur Holland Michel We can still have nice things A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.) + The secrets behind how Shazam works have been revealed.+ For the first time in a decade, a rare “Cloud Jaguar” was caught on camera.+ Explore

When generative AI first moved from research labs into real-world business applications, enterprises made a tacit bargain: “Capability now, control later.” Feed your proprietary data into third-party AI models, and you will get powerful results. But your data passes through systems you do not own, under governance you do not set. The protections you rely on are only as durable as the provider’s next policy update. Now, with generative AI established in everyday business operations and sophisticated new agentic AI systems advancing every day, companies are reevaluating the terms of that deal. “Data is really a new currency; it’s the IP for many companies,” says Kevin Dallas, CEO of EDB, echoing a recurrent anxiety from customers. “The big concern is, if you’re deploying an AI-infused application with a cloud-based large language model, are you losing your IP? Are you losing your competitive position?” DOWNLOAD THE REPORT That question is now fueling a movement toward reclaiming both the data and AI systems that have rapidly become part of core business infrastructure. AI and data sovereignty, which refers to breaking dependence on centralized providers and establishing genuine control over models and data estates, it is an urgent priority for many companies, says Dallas, citing internal EDB data: “70% of global executives believe they need a sovereign data and AI platform to be successful.” The idea of AI sovereignty is becoming a global policy conversation. NVIDIA CEO Jensen Huang recently spoke about the need for such a shift at the World Economic Forum’s annual meeting at Davos in January 2026: “I really believe that every country should get involved to build AI infrastructure, build your own AI, take advantage of your fundamental natural resource—which is your language and culture—develop your AI, continue to refine it, and have your national intelligence be part of your ecosystem.” This report explores how enterprises are pursuing sovereignty over their models and data estates in an era of rapid AI adoption. Drawing on a survey conducted by EDB of more than 2,050 senior executives and a series of interviews with industry experts, the research confirms that the sovereignty movement on the enterprise level is already well underway. Download the report. 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.