Latest AI News
From LLMs to hallucinations, here’s a simple guide to common AI terms
Artificial intelligence is a deep and convoluted world. The scientists who work in this field often rely on jargon and lingo to explain what they’re working on. As a result, we frequently have to use those technical terms in our coverage of the artificial intelligence industry. That’s why we thought it would be helpful to put together a glossary with definitions of some of the most important words and phrases that we use in our articles. We will regularly update this glossary to add new entries as researchers continually uncover novel methods to push the frontier of artificial intelligence while identifying emerging safety risks. Artificial general intelligence, or AGI, is a nebulous term. But it generally refers to AI that’s more capable than the average human at many, if not most, tasks. OpenAI CEO Sam Altmanrecentlydescribed AGI as the “equivalent of a median human that you could hire as a co-worker.” Meanwhile,OpenAI’s charterdefines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s understanding differs slightly from these two definitions; the lab views AGI as “AI that’s at least as capable as humans at most cognitive tasks.” Confused? Not to worry —so are experts at the forefront of AI research. An AI agent refers to a tool that uses AI technologies to perform a series of tasks on your behalf — beyond what a more basic AI chatbot could do — such as filing expenses, booking tickets or a table at a restaurant, or even writing and maintaining code. However, as we’veexplained before, there are lots of moving pieces in this emergent space, so “AI agent” might mean different things to different people. Infrastructure is also still being built out to deliver on its envisaged capabilities. But the basic concept implies an autonomous system that may draw on multiple AI systems to carry out multistep tasks. Given a simple question, a human brain can answer without even thinking too much about it — things like “which animal is taller, a giraffe or a cat?” But in many cases, you often need a pen and paper to come up with the right answer because there are intermediary steps. For instance, if a farmer has chickens and cows, and together they have 40 heads and 120 legs, you might need to write down a simple equation to come up with the answer (20 chickens and 20 cows). In an AI context, chain-of-thought reasoning for large language models means breaking down a problem into smaller, intermediate steps to improve the quality of the end result. It usually takes longer to get an answer, but the answer is more likely to be correct, especially in a logic or coding context. Reasoning models are developed from traditional large language models and optimized for chain-of-thought thinking thanks to reinforcement learning. (See:Large language model) Although somewhat of a multivalent term, compute generally refers to the vitalcomputational powerthat allows AI models to operate. This type of processing fuels the AI industry, giving it the ability to train and deploy its powerful models. The term is often a shorthand for the kinds of hardware that provides the computational power — things like GPUs, CPUs, TPUs, and other forms of infrastructure that form the bedrock of the modern AI industry. A subset of self-improving machine learning in which AI algorithms are designed with a multi-layered, artificial neural network (ANN) structure. This allows them to make more complex correlations compared to simpler machine learning-based systems, such as linear models or decision trees. The structure of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain. Deep learning AI models are able to identify important characteristics in data themselves, rather than requiring human engineers to define these features. The structure also supports algorithms that can learn from errors and, through a process of repetition and adjustment, improve their own outputs. However, deep learning systems require a lot of data points to yield good results (millions or more). They also typically take longer to train compared to simpler machine learning algorithms — so development costs tend to be higher. (See:Neural network) Diffusion is the tech at the heart of many art-, music-, and text-generating AI models. Inspired by physics,diffusion systems slowly “destroy” the structure of data— for example, photos, songs, and so on — by adding noise until there’s nothing left. In physics, diffusion is spontaneous and irreversible — sugar diffused in coffee can’t be restored to cube form. But diffusion systems in AI aim to learn a sort of “reverse diffusion” process to restore the destroyed data, gaining the ability to recover the data from noise. Distillation is a technique used to extract knowledge from a large AI model with a ‘teacher-student’ model. Developers send requests to a teacher model and record the outputs. Answers are sometimes compared with a dataset to see how accurate they are. These outputs are then used to train the student model, which is trained to approximate the teacher’s behavior. Distillation can be used to create a smaller, more efficient model based on a larger model with a minimal distillation loss. This is likely how OpenAI developed GPT-4 Turbo, a faster version of GPT-4. While all AI companies use distillation internally, it may have also been used by some AI companies to catch up with frontier models. Distillation from a competitor usuallyviolatesthe terms of service of AI API and chat assistants. This refers to the further training of an AI model to optimize performance for a more specific task or area than was previously a focal point of its training — typically by feeding in new, specialized (i.e., task-oriented) data. Many AI startups are taking large language models as a starting point to build a commercial product but are vying to amp up utility for a target sector or task by supplementing earlier training cycles with fine-tuning based on their own domain-specific knowledge and expertise. (See:Large language model [LLM]) A GAN, or Generative Adversarial Network, is a type of machine learning framework that underpins some important developments in generative AI when it comes to producing realistic data – including (but not only) deepfake tools. GANs involve the use of a pair of neural networks, one of which draws on its training data to generate an output that is passed to the other model to evaluate. This second, discriminator model thus plays the role of a classifier on the generator’s output – enabling it to improve over time. The GAN structure is set up as a competition (hence “adversarial”) – with the two models essentially programmed to try to outdo each other: the generator is trying to get its output past the discriminator, while the discriminator is working to spot artificially generated data. This structured contest can optimize AI outputs to be more realistic without the need for additional human intervention. Though GANs work best for narrower applications (such as producing realistic photos or videos), rather than general purpose AI. Hallucination is the AI industry’s preferred term for AI models making stuff up – literally generating information that is incorrect. Obviously, it’s a huge problem for AI quality. Hallucinations produce GenAI outputs that can be misleading and could even lead to real-life risks — with potentially dangerous consequences (think of a health query that returns harmful medical advice). This is why most GenAI tools’ small print now warns users to verify AI-generated answers, even though such disclaimers are usually far less prominent than the information the tools dispense at the touch of a button. The problem of AIs fabricating information is thought to arise as a consequence of gaps in training data. For general purpose GenAI especially — also sometimes known as foundation models — this looks difficult to resolve. There is simply not enough data in existence to train AI models to comprehensively resolve all the questions we could possibly ask. TL;DR: we haven’t invented God (yet). Hallucinations are contributing to a push towards increasingly specialized and/or vertical AI models — i.e. domain-specific AIs that require narrower expertise – as a way to reduce the likelihood of knowledge gaps and shrink disinformation risks. Inference is the process of running an AI model. It’s setting a model loose to make predictions or draw conclusions from previously seen data. To be clear, inference can’t happen without training; a model must learn patterns in a set of data before it can effectively extrapolate from this training data. Many types of hardware can perform inference, ranging from smartphone processors to beefy GPUs to custom-designed AI accelerators. But not all of them can run models equally well. Very large models would take ages to make predictions on, say, a laptop versus a cloud server with high-end AI chips. [See:Training] Large language models, or LLMs, are the AI models used by popular AI assistants, such asChatGPT,Claude,Google’s Gemini,Meta’s AI Llama,Microsoft Copilot, orMistral’s Le Chat. When you chat with an AI assistant, you interact with a large language model that processes your request directly or with the help of different available tools, such as web browsing or code interpreters. AI assistants and LLMs can have different names. For instance, GPT is OpenAI’s large language model and ChatGPT is the AI assistant product. LLMs are deep neural networks made of billions of numerical parameters (or weights, see below) that learn the relationships between words and phrases and create a representation of language, a sort of multidimensional map of words. These models are created from encoding the patterns they find in billions of books, articles, and transcripts. When you prompt an LLM, the model generates the most likely pattern that fits the prompt. It then evaluates the most probable next word after the last one based on what was said before. Repeat, repeat, and repeat. (See:Neural network) Memory cache refers to an important process that boosts inference (which is the process by which AI works to generate a response to a user’s query). In essence, caching is an optimization technique, designed to make inference more efficient. AI is obviously driven by high-octane mathematical calculations and every time those calculations are made, they use up more power. Caching is designed to cut down on the number of calculations a model might have to run by saving particular calculations for future user queries and operations. There are different kinds of memory caching, although one of the more well-known isKV (or key value) caching. KV caching works in transformer-based models, and increases efficiency, driving faster results by reducing the amount of time (and algorithmic labor) it takes to generate answers to user questions. (See:Inference) A neural network refers to the multi-layered algorithmic structure that underpins deep learning — and, more broadly, the whole boom in generative AI tools following the emergence of large language models. Although the idea of taking inspiration from the densely interconnected pathways of the human brain as a design structure for data processing algorithms dates all the way back to the 1940s, it was the much more recent rise of graphical processing hardware (GPUs) — via the video game industry — that really unlocked the power of this theory. These chips proved well suited to training algorithms with many more layers than was possible in earlier epochs — enabling neural network-based AI systems to achieve far better performance across many domains, including voice recognition, autonomous navigation, and drug discovery. (See:Large language model [LLM]) RAMageddon is the fun new term for a not-so-fun trend that is sweeping the tech industry: an ever-increasing shortage of random access memory, or RAM chips, which power pretty much all the tech products we use in our daily lives. As the AI industry has blossomed, the biggest tech companies and AI labs — all vying to have the most powerful and efficient AI — are buying so much RAM to power their data centers that there’s not much left for the rest of us. And that supply bottleneck means that what’s left is getting more and more expensive. That includes industries like gaming (where major companies have had toraise prices on consolesbecause it’s harder to find memory chips for their devices), consumer electronics (where memory shortage could causethe biggest dip in smartphone shipmentsin more than a decade), and general enterprise computing (because those companies can’t get enough RAM for their own data centers). The surge in prices is only expected to stop after the dreaded shortage ends but, unfortunately, there’snot really much of a signthat’s going to happen anytime soon. Developing machine learning AIs involves a process known as training. In simple terms, this refers to data being fed in in order that the model can learn from patterns and generate useful outputs. Things can get a bit philosophical at this point in the AI stack — since, pre-training, the mathematical structure that’s used as the starting point for developing a learning system is just a bunch of layers and random numbers. It’s only through training that the AI model really takes shape. Essentially, it’s the process of the system responding to characteristics in the data that enables it to adapt outputs towards a sought-for goal — whether that’s identifying images of cats or producing a haiku on demand. It’s important to note that not all AI requires training. Rules-based AIs that are programmed to follow manually predefined instructions — for example, such as linear chatbots — don’t need to undergo training. However, such AI systems are likely to be more constrained than (well-trained) self-learning systems. Still, training can be expensive because it requires lots of inputs — and, typically, the volumes of inputs required for such models have been trending upwards. Hybrid approaches can sometimes be used to shortcut model development and help manage costs. Such as doing data-driven fine-tuning of a rules-based AI — meaning development requires less data, compute, energy, and algorithmic complexity than if the developer had started building from scratch. [See:Inference] When it comes to human-machine communication, there are some obvious challenges. People communicate using human language, while AI programs execute tasks and respond to queries through complex algorithmic processes that are informed by data. In their simplest definition, tokens represent the basic building blocks of human-AI communication, in that they are discrete segments of data that have either been processed or produced by an LLM. Tokens are created via a process known as “tokenization,” which breaks down raw data and refines it into distinct units that are digestible to an LLM. Similar to how a software compiler translates human language into binary code that a computer can digest, tokenization interprets human language for an AI program via their user queries so that it can prepare a response. There are several different kinds of tokens — including input tokens (the kind that must be generated in response to a human user’s query), output tokens (the kind that are generated as the LLM responds to the human’s request), and reasoning tokens, which involve longer, more intensive tasks and processes that occur as part of a user request. With enterprise AI, token usage also determines costs. Since tokens are equivalent to the amount of data being processed by a model, they have also become the means by which the AI industry monetizes its services. Most AI companies charge for LLM usage on a per-token-basis. Thus, the more tokens a business burns as it uses an AI program (ChatGPT, for example), the more money it will have to pay its AI service provider (OpenAI). A technique where a previously trained AI model is used as the starting point for developing a new model for a different but typically related task – allowing knowledge gained in previous training cycles to be reapplied. Transfer learning can drive efficiency savings by shortcutting model development. It can also be useful when data for the task that the model is being developed for is somewhat limited. But it’s important to note that the approach has limitations. Models that rely on transfer learning to gain generalized capabilities will likely require training on additional data in order to perform well in their domain of focus (See:Fine tuning) Weights are core to AI training, as they determine how much importance (or weight) is given to different features (or input variables) in the data used for training the system — thereby shaping the AI model’s output. Put another way, weights are numerical parameters that define what’s most salient in a dataset for the given training task. They achieve their function by applying multiplication to inputs. Model training typically begins with weights that are randomly assigned, but as the process unfolds, the weights adjust as the model seeks to arrive at an output that more closely matches the target. For example, an AI model for predicting housing prices that’s trained on historical real estate data for a target location could include weights for features such as the number of bedrooms and bathrooms, whether a property is detached or semi-detached, whether it has parking, a garage, and so on. Ultimately, the weights the model attaches to each of these inputs reflect how much they influence the value of a property, based on the given dataset. This article is updated regularly with new information.
View
Snowflake’s 9,100 AI Customers Signal the Real Indian Enterprise Shift
Snowflake is not competing with model providers. It is ensuring that whatever model an enterprise uses, the output is reliable.
View
Sam Altman responds to ‘incendiary’ New Yorker article after attack on his home
OpenAI CEO Sam Altmanpublished a blog poston Friday evening responding to both an apparent attack on his home andan in-depth New Yorker profileraising questions about his trustworthiness. Early Friday morning, someone allegedly threw a Molotov cocktail at Altman’s San Francisco home. No one was hurt in the incident, and a suspect was later arrested at OpenAI headquarters, where he was threatening to burn down the building,according to the SF Police Department. While the police have not identified the suspect publicly, Altman noted that the incident came a few days after “an incendiary article” was published about him. He said someone had suggested that the article’s publication “at a time of great anxiety about AI” could make things “more dangerous” for him. “I brushed it aside,” Altman said. “Now I am awake in the middle of the night and pissed, and thinking that I have underestimated the power of words and narratives.” The article in question was a lengthy investigative piece written by Ronan Farrow (who won a Pulitzer for reporting that revealed many of the sexual abuse allegations around Harvey Weinstein) and Andrew Marantz (who’s written extensively about technology and politics). Farrow and Marantz said that during interviews with more than 100 people who have knowledge of Altman’s business conduct, most described Altman as someone with “a relentless will to power that, even among industrialists who put their names on spaceships, sets him apart.” Echoingother journalists who have profiled Altman, Farrow and Marantz suggested that many sources raised questions about his trustworthiness, with one anonymous board member saying he combines “a strong desire to please people, to be liked in any given interaction” with “a sociopathic lack of concern for the consequences that may come from deceiving someone.” In his response, Altman said that looking back, he can identify “a lot of things I’m proud of and a bunch of mistakes.” Among the mistakes, he said, is a tendency towards “being conflict-averse,” which he said has “caused great pain for me and OpenAI.” “I am not proud of handling myself badly in a conflict with our previous board that led to a huge mess for the company,” Altman said, presumably referring tohis removal and rapid reinstatement as OpenAI CEOback in 2023. “I have made many other mistakes throughout the insane trajectory of OpenAI; I am a flawed person in the center of an exceptionally complex situation, trying to get a little better each year, always working for the mission.” He added, “I am sorry to people I’ve hurt and wish I had learned more faster.” Altman also acknowledged that there seems to be “so much Shakespearean drama between the companies in our field,” which he attributed to a “‘ring of power’ dynamic” that “makes people do crazy things.” Of course, the correct way to deal with thering of poweris to destroy it, so Altman added, “I don’t mean that [artificial general intelligence] is the ring itself, but instead the totalizing philosophy of ‘being the one to control AGI.’” His proposed solution is “to orient towards sharing the technology with people broadly, and for no one to have the ring.” Altman concluded by saying that he welcomes “good-faith criticism and debate,” while reiterating his belief that “technological progress can make the future unbelievably good, for your family and mine.” “While we have that debate, we should de-escalate the rhetoric and tactics and try to have fewer explosions in fewer homes, figuratively and literally,” he said.
View
India Hits 27 Million Developers on GitHub
Over 2 million developers joined GitHub from India this year.
View
India’s AI Dream Falls Short of a Million GPUs
Without scaling GPUs and investment, India risks trailing in the frontier AI race.
View
Claude Mythos Explained: Everything You Need to Know About Anthropic’s Cybersecurity AI Model
In just 48 hours, Anthropic announced its new cybersecurity-focused artificial intelligence (AI) model, Claude Mythos Preview, and raised alarms across the entire global tech space. The San Francisco-based AI startup called it the most powerful model when it comes to cybersecurity tasks, especially finding undiscovered vulnerabilities in codebases. The company also warned that the model found thousands of high-severity vulnerabilities in “every major operating system and web browser,” which, if true, is a major concern. Anthropic has also limited its release, citing its ability to hack into any system.
View
Anthropic temporarily banned OpenClaw’s creator from accessing Claude
“Yeah folks, it’s gonna be harder in the future to ensure OpenClaw still works with Anthropic models,” OpenClaw creator Peter Steinbergerposted on X early Friday morning, along with a photo of a message from Anthropic saying his account had been suspended over “suspicious” activity. The ban didn’t last long. A few hours later, after the post went viral, Steinberger said his account had been reinstated. Among hundreds of comments — many of them in conspiracy theory land, given that Steinberger isnow employedby Anthropic rival OpenAI — was one by an Anthropic engineer. The engineer told the famed developer that Anthropic has never banned anyone for using OpenClaw and offered to help. Yeah folks, it's gonna be harder in the future to ensure OpenClaw still works with Anthropic models.pic.twitter.com/U6F8GZvPcH It’s not clear if that was the key that restored the account. (We’ve asked Anthropic about it.) But the whole message string was enlightening on many levels. To recap the recent history: This ban followed news last week thatsubscriptions to Anthropic’s Claude would no longer cover“third-party harnesses including OpenClaw,” the AI model company said. OpenClaw users now have to pay for that usage separately, based on consumption, through Claude’s API. In essence, Anthropic, which offers its own agent, Cowork, is now charging a “claw tax.” Steinberger said he was following this new rule and using his API but was banned anyway. Anthropic said it instituted the pricing change because subscriptions weren’t built to handle the “usage patterns” of claws. Claws can bemore compute-intensivethan prompts or simple scripts because they may run continuous reasoning loops, automatically repeat or retry tasks, and tie into a lot of other third-party tools. Steinberger, however, wasn’t buying that excuse. After Anthropic changed the pricing,he posted, “Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.” Though he didn’t specify, he may have been referring to features added to Claude’s Cowork agent, such as Claude Dispatch,which lets users remotely control agents and assign tasks. Dispatch rolled out a couple of weeks before Anthropic changed its OpenClaw pricing policy. Steinberger’s frustration with Anthropic was again on display Friday. One person implied that some of this is on him for taking a job at OpenAI instead of Anthropic, posting, “You had the choice, but you went to the wrong one.” To which Steinberger replied: “One welcomed me, one sent legal threats.” Ouch. When multiple people asked him why he’s using Claude instead of his employer’s models at all, he explained that he only uses it for testing, to ensure updates to OpenClaw won’t break things for Claude users. He explained: “You need to separate two things. My work at the OpenClaw Foundation where we wanna make OpenClaw work great for *any* model provider, and my job at OpenAI to help them with future product strategy.” Multiple people also pointed out that the need to test Claude is because that model remains a popular choice for OpenClaw users over ChatGPT. He also heard that when Anthropic changed its pricing, to which he replied: “Working on that.” (So, that’s a clue about what his job at OpenAI entails.) Steinberger did not respond to a request for comment.
View
TechCrunch is heading to Tokyo — and bringing the Startup Battlefield with it
TechCrunch is partnering withSusHi Tech Tokyo 2026, Asia’s largest global innovation conference, taking place April 27–29 at Tokyo Big Sight. And we’re not just showing up to cover it — our very own Startup Battlefield program manager, Isabelle Johannessen, will be on the ground as a judge for the SusHi Tech Challenge, the conference’s flagship global pitch competition. For the winner, the stakes couldn’t be higher: The SusHi Tech Challenge Grand Prix recipient will be automatically entered into the TechCrunch Disrupt Startup Battlefield Top 200 — making them eligible to pitch on one of the most coveted stages in the startup world. Now in its fourth year,SusHi Tech Tokyo— short forSustainableHigh City Tech Tokyo — has grown into the largest innovation conference in Asia, drawing startups, investors, corporate partners, and city leaders from around the world. This year’s edition is the biggest yet: 750 startup exhibitors from 60 countries, more than 10,000 facilitated business meetings, and an expected 60,000 attendees across three days. The conference is organized by the Tokyo Metropolitan Government with a clear mission: bring together the world’s best innovators to build the sustainable cities of the future. On the expo floor, 62 corporate partners — including Sony, Google, Microsoft, and Mizuho — are hosting reverse pitches and actively seeking startup collaborators, making it as much a live dealmaking marketplace as a conference. And the programming reflects that ambition. SusHi Tech 2026 is zeroing in on four technology domains reshaping society: AI, Robotics, Resilience, and Entertainment. Expect live demos of humanoid robots, sessions on autonomous driving’s software revolution, deep dives into cyber defense and climate tech, and candid conversations about how AI is rewriting the global music and anime industries. Speakers include Howard Wright (Nvidia), Rob Chu (AWS), Eva Chen (Trend Micro), Qasar Younis (Applied Intuition), Christine Tsai (500 Global), Kathy Matsui (MPower Partners), and Tokyo governor Yuriko Koike, among many others. Roughly 60% of speakers come from outside Japan, and approximately half are women. Going to be in Tokyo? Don’t miss it.Get your tickets here. The pitch competition drew 820 applications from 60 countries and regions — 437 international, 383 Japanese. Twenty semifinalists compete on April 27, seven finalists advance to the final on April 28, and one Grand Prix winner takes home ¥10,000,000 and automatic entry into the TechCrunch Disrupt Startup Battlefield Top 200. The conference extends well beyond the convention floor. City leaders from 49 cities across five continents — from Los Angeles to Nairobi to Singapore — are convening for the G-NETS Leaders Summit to forge concrete commitments on climate resilience and urban sustainability. On the expo floor, 62 corporate partners, including Sony, Google, Microsoft, and Mizuho, are hosting reverse pitches and actively seeking startup collaborators. And because this is Tokyo, the experience doesn’t stop at 6 p.m.: Classical music performances from La Folle Journée, waterfront cruises along Tokyo Bay, and the Tokyo Innovation NIGHTs networking series round out the program. The officialSusHi Tech Tokyo 2026 Official appis your command center on the ground. Before you even arrive, AI-powered matching recommends the right startups, investors, and partners for you to connect with — and lets you book meeting rooms in advance. On-site, a GPS floor map, QR business card exchange, and real-time push notifications keep you oriented across the sprawling Tokyo Big Sight venue. Download foriOSorAndroid. SusHi Tech Tokyo 2026runs April 27–29 at Tokyo Big Sight. Business days are April 27–28; Public Day (free admission) is April 29.
View
Stalking victim sues OpenAI, claims ChatGPT fueled her abuser’s delusions and ignored her warnings
After months of conversations with ChatGPT, a 53-year-old Silicon Valley entrepreneur became convinced he’d discovered a cure for sleep apnea and that powerful people were coming after him, according to a new lawsuit filed in California Superior Court in San Francisco County. He then allegedly used the tool to stalk and harass his ex-girlfriend. Now the ex-girlfriend is suing OpenAI, alleging the company’s technology enabled the acceleration of her harassment, TechCrunch has exclusively learned. She claims OpenAI ignored three separate warnings that the user posed a threat to others, including an internal flag classifying his account activity as involving mass-casualty weapons. The plaintiff, referred to as Jane Doe to protect her identity, is suing for punitive damages. She also filed a temporary restraining order Friday asking the court to force OpenAI to block the user’s account, prevent him from creating new ones, notify her if he attempts to access ChatGPT, and preserve his complete chat logs for discovery. OpenAI has agreed to suspend the user’s account but has refused the rest, according to Doe’s lawyers. They say the company is withholding information about specific plans for harming Doe and other potential victims the user may have discussed with ChatGPT. The lawsuit lands amid growing concern over the real-world risks of sycophantic AI systems. GPT-4o, the model cited in this and many other cases, wasretired from ChatGPT in February. The case is brought by Edelson PC, the firm behind the wrongful death suits involving teenagerAdam Raine, who died by suicide after months of conversations with ChatGPT, andJonathan Gavalas, whose family alleges Google’s Gemini fueled his delusions and potential mass-casualty event before his death. Lead attorney Jay Edelson has warned that AI-induced psychosis is escalating fromindividual harm toward mass-casualty events. That legal pressure is now colliding directly with OpenAI’s legislative strategy: The company isbacking an Illinois billthat would shield AI labs from liability even in cases involving mass deaths or catastrophic financial harm. OpenAI did not respond in time to comment. TechCrunch will update the article if the company responds. The Jane Doe lawsuit lays out in detail how that liability played out for one woman over several months. Last year, the ChatGPT user in the lawsuit (whose name is not included in the lawsuit to protect his identity) became convinced that he had invented a cure for sleep apnea after months of “high volume, sustained use of GPT-4o.” When no one took his work seriously, ChatGPT told him that “powerful forces” were watching him, including using helicopters to surveil his activities, according to the complaint. In July 2025, Jane Doe urged him to stop using ChatGPT and to seek help from a mental health professional. He instead turned back to ChatGPT, which assured him he was “a level 10 in sanity” and helped him double down on his delusions, per the lawsuit. Doe had broken up with the user in 2024, and he used ChatGPT to process the split, according to emails and communications cited in the lawsuit. Rather than push back on his one-sided account, it repeatedly cast him as rational and wronged, and her as manipulative and unstable. He then took these AI-generated conclusions off the screen and into the real world, using them to stalk and harass her. This manifested in several AI-generated, clinical-looking psychological reports that he distributed to her family, friends, and employer. Meanwhile, the user continued to spiral. In August 2025, OpenAI’s automated safety system flagged him for “Mass Casualty Weapons” activity and deactivated his account. A human safety team member reviewed the account the next day and restored it, even though his account may have contained evidence that he was targeting and stalking individuals, including Doe, in real life. For example, a September screenshot the user sent to Doe showed a list of conversation titles including “violence list expansion” and “fetal suffocation calculation.” The decision to reinstate is notable following two recent school shootings in Tumbler Ridge, Canada, and at Florida State University (FSU). OpenAI’s safety team had flagged the Tumbler Ridge shooter as a potential threat, but higher-upsreportedlydecided not to alert authorities. Florida’s attorney generalthis week opened an investigationinto OpenAI’s possible link with the FSU shooter. According to the Jane Doe lawsuit, when OpenAI restored her stalker’s account, his Pro subscription wasn’t reinstated alongside it. He emailed the trust and safety team to sort it out, copying Doe on the message. In his emails, he wrote things like: “I NEED HELP VERY FAST, PLEASE. PLEASE CALL ME!” and “this is a matter of life or death.” He claimed he was “in the process of writing 215 scientific papers,” which he was writing so fast he didn’t “even have time to read.” Included in those emails was a list of tens of AI-generated “scientific papers” with titles like: “Deconstructing Race as a Biological Category_ Legal, Scientific, and Horn of Africa Perspectives.pdf.txt.” “The user’s communications provided unmistakable notice that he was mentally unstable and that ChatGPT was the engine of his delusional thinking and escalating conduct,” the lawsuit states. “The user’s stream of urgent, disorganized, and grandiose claims, along with a concrete ChatGPT-generated report targeting Plaintiff by name and a sprawling body of purported ‘scientific’ materials, was unmistakable evidence of that reality. OpenAI did not intervene, restrict his access, or implement any safeguards. Instead, it enabled him to continue using the account and restored his full Pro access.” Doe, who claims in the lawsuit that she was living in fear and could not sleep in her own home, submitted a Notice of Abuse to OpenAI in November. “For the last seven months, he has weaponized this technology to create public destruction and humiliation against me that would have been impossible otherwise,” Doe wrote in her letter to OpenAI requesting the company permanently ban the user’s account. OpenAI responded, acknowledging the report was “extremely serious and troubling” and that it was carefully reviewing the information. Doe never heard back. Over the next couple of months, the user continued to harass Doe, sending her a series of threatening voicemails. In January, he was arrested and charged with four felony counts of communicating bomb threats and assault with a deadly weapon. Doe’s lawyers allege this validates warnings both she and OpenAI’s own safety systems had raised months earlier, warnings the company allegedly chose to ignore. The user was found incompetent to stand trial and committed to a mental health facility, but a “procedural failure by the State” means he will soon be released to the public, according to Doe’s lawyers. Edelson called on OpenAI to cooperate. “In every case, OpenAI has chosen to hide critical safety information — from the public, from victims, from people its product is actively putting in danger,” he said. “We’re calling on them, for once, to do the right thing. Human lives must mean more than OpenAI’s race to an IPO.”
View
Last 24 hours: Save up to $500 on your TechCrunch Disrupt 2026 pass
This is it. The clock is running out. Tonight is your last chance to lock in savings of up to $500 for yourTechCrunch Disrupt 2026pass. These discounts end at 11:59 p.m. PT. Register here to secure yours with the limited-time offer. This year,Disrupttakes over San Francisco’s Moscone West from October 13–15, bringing together 10,000 founders, VCs, operators, and tech leaders for a tightly curated, three-day experience focused on real outcomes. Attendees return for: With300+ startupsexpected to showcase their innovations across the venue, the intensity of the live pitch competitionStartup Battlefield 200, and curated networking designed to drive results, Disrupt isn’t just another conference. It’s where momentum is built. Disrupt isn’t about wandering between sessions. It’s about intentional connections and curated experiences designed for how people actually grow in tech. If you’re hands-on in tech, Disrupt was built for you. Founders meet investors actively backing breakthrough ideas. VCs cut through the noise to discover startups aligned with their investment focus. Operators exchange real-world lessons on building, scaling, and shipping what’s next. Aspiring innovators get a front-row seat to tomorrow’s tech. Each Disrupt brings together 250+ of the most influential names in tech, leaders who have shaped the industry and continue to define what’s next. Keep an eye on theDisrupt 2026 event pageas the agenda goes live to see who will take the stage this year. Past speakers include: At 11:59 p.m. PT tonight, prices go up and this opportunity will be gone. Disrupt will still be filled with the same founders, investors, and operators you’ll meet. The only difference is what you paid to be there. If Disrupt is part of your 2026 strategy, make the move now. Secure your pass, lock in the savings, and step into the conversations that move your business forward.Register before today ends.
View
India's Data Centre Boom Outpaces Power Planning
As data centres expand rapidly, rising energy demand strains grids and threatens India’s clean-power ambitions, warns a parliamentary panel.
View
Google Rolls Out AI Mode Agentic Features in India, Enables Restaurant Booking via Search
Google on Friday announced the rollout of agentic capabilities in AI Mode in India. The update introduces the ability to discover and book restaurant reservations directly through Search. As per the Mountain View-based tech giant, it is aimed at achieving a practical use case by helping users complete real-world tasks more efficiently. Google claims the new agentic capabilities can handle multi-step queries and reduce the effort required to search across platforms manually.
View