vLLM is an open-source software framework for inference and serving of large language models and related multimodal models. Originally developed at the University of California, Berkeley's Sky Computing Lab, the project is centered on PagedAttention, a memory-management method for transformer key–value caches, and supports features such as continuous batching, distributed inference, quantization, and OpenAI-compatible APIs. According to a project maintainer, the "v" in vLLM originally referred to "virtual", inspired by virtual memory. == History == vLLM was introduced in 2023 by researchers affiliated with the Sky Computing Lab at UC Berkeley. Its core ideas were described in the 2023 paper Efficient Memory Management for Large Language Model Serving with PagedAttention, which presented the system as a high-throughput and memory-efficient serving engine for large language models. In 2025, the PyTorch Foundation announced that vLLM had become a Foundation-hosted project. PyTorch's project page states that the University of California, Berkeley contributed vLLM to the Linux Foundation in July 2024. In January 2026, TechCrunch reported that the creators of vLLM had launched the startup Inferact to commercialize the project, raising $150 million in seed funding. == Architecture == According to its 2023 paper, vLLM was designed to improve the efficiency of large language model serving by reducing memory waste in the key–value cache used during transformer inference. The paper introduced PagedAttention, an algorithm inspired by virtual memory and paging techniques in operating systems, and described vLLM as using block-level memory management and request scheduling to increase throughput while maintaining similar latency. The project documentation and repository describe support for continuous batching, chunked prefill, speculative decoding, prefix caching, quantization, and multiple forms of distributed inference and serving. PyTorch has described vLLM as a high-throughput, memory-efficient inference and serving engine that supports a range of hardware back ends, including NVIDIA and AMD GPUs, Google TPUs, AWS Trainium, and Intel processors.
Camfecting
In computer security, camfecting is the process of attempting to hack into a person's webcam and activate it without the webcam owner's permission. The remotely activated webcam can be used to watch anything within the webcam's field of vision, sometimes including the webcam owner themselves. Camfecting is most often carried out by infecting the victim's computer with a virus that can provide the hacker access to their webcam. This attack is specifically targeted at the victim's webcam, and hence the name camfecting, a portmanteau of the words camera and infecting. Typically, a webcam hacker or a camfecter sends his victim an innocent-looking application which has a hidden Trojan software through which the camfecter can control the victim's webcam. The camfecter virus installs itself silently when the victim runs the original application. Once installed, the camfecter can turn on the webcam and capture pictures/videos. The camfecter software works just like the original webcam software present in the victim computer, the only difference being that the camfecter controls the software instead of the webcam's owner. == Notable cases == Marcus Thomas, former assistant director of the FBI's Operational Technology Division in Quantico, said in a 2013 story in The Washington Post that the FBI had been able to covertly activate a computer's camera—without triggering the light that lets users know it is recording—for several years. In November 2013, American teenager Jared James Abrahams pleaded guilty to hacking over 100-150 women and installing the highly invasive malware Blackshades on their computers in order to obtain nude images and videos of them. One of his victims was Miss Teen USA 2013 Cassidy Wolf. Researchers from Johns Hopkins University have shown how to covertly capture images from the iSight camera on MacBook and iMac models released before 2008, by reprogramming the microcontroller's firmware. == Prevention == A computer that does not have an up-to-date webcam software or any anti-virus (or firewall) software installed and operational may be at increased risk for camfecting from different types of malware. Softcams may nominally increase this risk, if not maintained or configured properly. Although a person cannot protect themselves from zero-day exploits that could potentially activate a camera unknowingly, such as Pegasus is able to do on smartphones. The only way to truly avoid being watched through your own camera is by blocking it physically, since software blocks can be overriden by advanced persistent threats. A simple piece of tape is more commonly used to offuscate the feed of the camera. With even Mark Zuckerberg doing so on his personal laptop that appeared during a presentation. And it being the way Snowden, an ex-contractor for the NSA, is portrayed to do so to prevent camfecting in the biopic Snowden. There is now a market for the manufacture and sale of sliding lens covers that allow users to physically block their computer's camera and, in some cases, microphone. A number of phone and laptop manufacturers tried to implement pop-up cameras that can only be opened manually by the user. But the trend did not become mainstream because of the engineering it took to keep the mechanisms up to date, aswell as the fragility and durability of the cameras.
Sarvam AI
Sarvam AI is an Indian artificial intelligence company headquartered in Bengaluru, Karnataka. Founded in 2023, the company develops large language models (LLMs) and multimodal AI systems with a focus on Indian languages and region-specific use cases. The company has received venture capital backing and has participated in government-supported AI initiatives, including India's sovereign large language model programme under the IndiaAI Mission. == History == Sarvam AI was founded in August 2023 by Vivek Raghavan and Pratyush Kumar, who were previously associated with AI4Bharat at the Indian Institute of Technology Madras. In December 2023, the company announced a combined seed and Series A funding round of approximately US$41 million. The round was led by Lightspeed Venture Partners, with participation from Peak XV Partners and Khosla Ventures. In April 2025, the Ministry of Electronics and Information Technology (MeitY) selected Sarvam AI as one of the companies to develop an indigenous foundational model under the IndiaAI Mission. As part of the initiative, the company received access to government-supported computing infrastructure, including GPUs allocated for model training over a specified period. In February 2026, Sarvam AI introduced two large language models at the AI Impact Summit held at Bharat Mandapam, New Delhi. == Products and technology == Sarvam AI develops language models trained on datasets that include multiple Indian languages and code-mixed text. The company uses mixture-of-experts (MoE) architectures in some of its models. === Foundational language models === On 18 February 2026, the company announced the release of two foundational models: Sarvam-30B – A 30-billion parameter model based on a mixture-of-experts design. According to company disclosures reported by the media, the model activates approximately 1 billion parameters per token and supports a 32,000-token context window. Sarvam-105B – A 105-billion parameter model activating approximately 9 billion parameters per token, with a 128,000-token context window. The model is positioned for complex reasoning and enterprise applications. On 20th February 2026, the company released a beta version of the Sarvam-105B model which is named Indus. It is available on the Apple App Store, Google Play Store and the web. === Speech and vision systems === Sarvam AI has also developed multimodal systems including speech-to-text and vision-language models. Its speech model, referred to as Saaras V3 in company materials, supports multiple Indian languages. The company has also introduced a vision-language model known as Sarvam Vision, intended for document understanding and optical character recognition (OCR) in Indian scripts. === Devices === 'Sarvam Kaze' is an indigenous AI-powered wearable glass that listens, understands, and captures what users see the world through their eyes in real time. The device supports more than 10 Indian languages, enabling voice-based interaction and potentially real-time translation. The company plans to launch the device in May 2026. == Startup support == In March 2026, Sarvam AI launched the Sarvam Startup Program, an initiative providing selected early-stage companies with 6–12 months of API credits scaled to their needs, priority engineering support, and access to production infrastructure for developing multilingual AI applications in areas such as speech, translation, and large language models. == Open-source release == In February 2026, Sarvam AI announced and open-sourced two large language models: Sarvam 30B (30 billion parameters) and Sarvam 105B (105 billion parameters, using a Mixture-of-Experts architecture with 10.3 billion active parameters). Both models were trained from scratch on datasets focused on Indian languages and support advanced reasoning, multilingual tasks, mathematics, and coding. The models are hosted on Hugging Face under the Apache License and are intended for enterprise and developer applications in Indian languages. The models were subsequently released as open source under the Apache License 2.0, with model weights made available on Hugging Face (sarvamai/sarvam-30b and sarvamai/sarvam-105b) and AIKosh in early March 2026. == Government and institutional collaborations == In 2025, Sarvam AI was selected to contribute to India's sovereign AI model initiative under the IndiaAI Mission. The initiative aims to support domestic AI infrastructure and model development. In March 2025, the Unique Identification Authority of India (UIDAI) announced a collaboration with Sarvam AI to integrate AI-based voice interactions and multilingual support into Aadhaar-related services. Sarvam AI has also worked with AI4Bharat and academic institutions on language datasets and speech research projects. == Industry participation == Sarvam AI presented its foundational models at the India AI Impact Summit 2026 in New Delhi. The company has also been listed among Indian members of the AI Alliance, a consortium focused on open-source artificial intelligence initiatives. == List of models ==
Liveness test
A liveness test, liveness check or liveness detection is an automated method for determining whether a subject is a real person or part of a spoofing attack. The technique is used as part of know your customer checks in financial services and during facial age estimation. Liveness detection is a cornerstone of digital safety. == Test process == The threat in face spoofing attacks is that "the attacker only needs to find a good face swap library on Github and understand how to inject the model into the camera feed during the KYC process". Fraudsters usually buy stolen IDs on the dark web to start a deepfake attack. An AI-powered generative adversarial network (GAN) can then generate the face swapping model that many online verification services fail to detect. Low level hackers may use face swapping apps such as SwapFace, DeepFaceLive, and Swapstream (increasing interest for those apps in 2023 according to Google Trends). In a video liveness test, users are typically asked to look into a camera and to move, smile or blink, and features of their moving face may then be compared to that of a still image. Artificial intelligence is used to counter presentation attacks such as deepfakes or users wearing hyperrealistic masks, or video injection attacks. Other forms of liveness test include checking for a pulse when using a fingerprint scanner or checking that a person's voice is not a recording or artificially generated during speaker recognition. == Adoption and certification == In a 2022 report published by the security firm Sensity, it was demonstrated that the liveness test of most US banks was easily cheated with new and publicly-available AI-powered techniques. Many of these banks disregarded the results of the report. In the first half of 2023, the security firm iProov detected a 704% increase in face-swap attacks. In 2023, in the UK, many customers of Ryanair were upset to have to go through many ID verification checks, including liveness tests, before boarding, as the airline was using it as a mean to deter customers to buy tickets through third-party websites. In the first half of 2024 iBeta Quality Assurance issued 18 new ISO/IEC 30107-3 Presentation Attack Detection certificates, raising the cumulative total to 85 since 2018. In January 2024, the Department of Homeland Security (DHS) opened applications from vendors to test their Liveness test. Identity frauds peaked during the COVID-19 lockdown, leading government agencies to take reinforced measures to secure their digital applications.
GPT-5.3-Codex
GPT-5.3-Codex (Generative Pre-trained Transformer 5.3 Codex) is a large language model (LLM) announced and released by OpenAI on February 5, 2026. It is made as a competitor to Claude's Opus 4.6, focusing on code generation, speed and the ability to search repositories, run terminal commands and at the same time, debug code. In technical benchmarks, it is reported that GPT-5.3 Codex is 25% faster than Opus 4.6. GPT-5.3 Codex is available in the Codex app and on the web; access via API is also planned. According to OpenAI, GPT-5.3-Codex is the company's "first model that was instrumental in creating itself." On February 12, 2026, GPT-5.3-Codex-Spark was released in a research preview, which is a smaller version of GPT-5.3-Codex which supports text-only input. As of February 2026, GPT-5.3-Codex is only available for ChatGPT Pro ($200/month) subscribers.
AI alignment
In the field of artificial intelligence (AI), alignment aims to steer AI systems toward a person's or group's intended goals, preferences, or ethical principles. An AI system is considered aligned if it advances the intended objectives. A misaligned AI system pursues unintended objectives. It is often difficult for AI designers to specify the full range of desired and undesired behaviors. Therefore, the designers often use simpler proxy goals, such as gaining human approval. But proxy goals can overlook necessary constraints or reward the AI system for merely appearing aligned. AI systems may also find loopholes that allow them to accomplish their proxy goals efficiently but in unintended, sometimes harmful, ways (reward hacking). Advanced AI systems may develop unwanted instrumental strategies, such as seeking power or self-preservation because such strategies help them achieve their assigned final goals. Furthermore, they might develop undesirable emergent goals that could be hard to detect before the system is deployed and encounters new situations and data distributions. Empirical research showed in 2024 that advanced large language models (LLMs) such as OpenAI o1 or Claude 3 sometimes engage in strategic deception to achieve their goals or prevent them from being changed. Some of these issues affect existing commercial systems such as LLMs, robots, autonomous vehicles, and social media recommendation engines. Some AI researchers argue that more capable future systems will be more severely affected because these problems partially result from high capabilities. Many prominent AI researchers and AI company leaders have argued or asserted that AI is approaching human-like (AGI) and superhuman cognitive capabilities (ASI), and could endanger human civilization if misaligned. These include "AI godfathers" Geoffrey Hinton and Yoshua Bengio and the CEOs of OpenAI, Anthropic, and Google DeepMind. These risks remain debated. AI alignment is a subfield of AI safety, the study of how to build safe AI systems. Other subfields of AI safety include robustness, monitoring, and capability control. Research challenges in alignment include instilling complex values in AI, developing honest AI, scalable oversight, auditing and interpreting AI models, and preventing emergent AI behaviors like power-seeking. Alignment research has connections to interpretability research, (adversarial) robustness, anomaly detection, calibrated uncertainty, formal verification, preference learning, safety-critical engineering, game theory, algorithmic fairness, and social sciences. == Objectives in AI == Programmers provide an AI system such as AlphaZero with an "objective function", in which they intend to encapsulate the goal(s) the AI is configured to accomplish. Such a system later populates a (possibly implicit) internal "model" of its environment. This model encapsulates all the agent's beliefs about the world. The AI then creates and executes whatever plan is calculated to maximize the value of its objective function. For example, when AlphaZero is trained on chess, it has a simple objective function of "+1 if AlphaZero wins, −1 if AlphaZero loses". During the game, AlphaZero attempts to execute whatever sequence of moves it judges most likely to attain the maximum value of +1. Similarly, a reinforcement learning system can have a "reward function" that allows the programmers to shape the AI's desired behavior. An evolutionary algorithm's behavior is shaped by a "fitness function". == Alignment problem == In 1960, AI pioneer Norbert Wiener described the AI alignment problem as follows: If we use, to achieve our purposes, a mechanical agency with whose operation we cannot interfere effectively [...] we had better be quite sure that the purpose put into the machine is the purpose which we really desire. AI alignment refers to ensuring that an AI system's objectives match some target. The target is variously defined as the goals of the system's designers or users, widely shared values, objective ethical standards, legal requirements, or the intentions its designers would have if they were more informed and enlightened. In democratic AI alignment, the target is the values and preferences of median voters, which increases political legitimacy. AI alignment is an open problem for modern AI systems and is a research field within AI. Aligning AI involves two main challenges: carefully specifying the purpose of the system (outer alignment) and ensuring that the system adopts the specification robustly (inner alignment). Researchers also attempt to create AI models that have robust alignment, sticking to safety constraints even when users adversarially try to bypass them. === Specification gaming and side effects === To specify an AI system's purpose, AI designers typically provide an objective function, examples, or feedback to the system. But designers are often unable to completely specify all important values and constraints, so they resort to easy-to-specify proxy goals such as maximizing the approval of human overseers, who are fallible. As a result, AI systems can find loopholes that help them accomplish the specified objective efficiently but in unintended, possibly harmful ways. This tendency is known as specification gaming or reward hacking, and is an instance of Goodhart's law. As AI systems become more capable, they are often able to game their specifications more effectively. Specification gaming has been observed in numerous AI systems. OpenAI GPT models for programming—including in real-world cases—have been found to explicitly plan hacking the tests used to evaluate them to falsely appear successful (e.g., explicitly stating "let's hack"). When the company penalized this, many models learned to obfuscate their plans while continuing to hack the tests. Another system was trained to finish a simulated boat race by rewarding the system for hitting targets along the track, but the system achieved more reward by looping and crashing into the same targets indefinitely. A 2025 Palisade Research study found that when tasked to win at chess against a stronger opponent, some reasoning LLMs attempted to hack the game system, for example by modifying or entirely deleting their opponent. Some alignment researchers aim to help humans detect specification gaming and steer AI systems toward carefully specified objectives that are safe and useful to pursue. When a misaligned AI system is deployed, it can have consequential side effects. Social media platforms have been known to optimize their recommendation algorithms for click-through rates, causing user addiction on a global scale. Stanford researchers say that such recommender systems are misaligned with their users because they "optimize simple engagement metrics rather than a harder-to-measure combination of societal and consumer well-being". Explaining such side effects, Berkeley computer scientist Stuart J. Russell said that the omission of implicit constraints can cause harm: "A system [...] will often set [...] unconstrained variables to extreme values; if one of those unconstrained variables is actually something we care about, the solution found may be highly undesirable. This is essentially the old story of the genie in the lamp, or the sorcerer's apprentice, or King Midas: you get exactly what you ask for, not what you want." Some researchers suggest that AI designers specify their desired goals by listing forbidden actions or by formalizing ethical rules (as with Asimov's Three Laws of Robotics). But Russell and Norvig argue that this approach overlooks the complexity of human values: "It is certainly very hard, and perhaps impossible, for mere humans to anticipate and rule out in advance all the disastrous ways the machine could choose to achieve a specified objective." Additionally, even if an AI system fully understands human intentions, it may still disregard them, because following human intentions may not be its objective (unless it is already fully aligned). === Pressure to deploy unsafe systems === Commercial organizations sometimes have incentives to take shortcuts on safety and to deploy misaligned or unsafe AI systems. For example, social media recommender systems have been profitable despite creating unwanted addiction and polarization. Competitive pressure can also lead to a race to the bottom on AI safety standards. For example, OpenAI has been sued for releasing a ChatGPT version that encouraged suicide for some unstable users, a behavior the company had overlooked amid a rushed product release. Similarly, in 2018, a self-driving car killed a pedestrian (Elaine Herzberg) after engineers disabled the emergency braking system because it was oversensitive and slowed development. === Risks from advanced misaligned AI === Some researchers are interested in aligning increasingly advanced AI systems, as progress in AI development is rapid, and industry and governments are trying to build advan
Minimum intelligent signal test
The minimum intelligent signal test, or MIST, is a variation of the Turing test proposed by Chris McKinstry in which only boolean (yes/no or true/false) answers may be given to questions. The purpose of such a test is to provide a quantitative statistical measure of humanness, which may subsequently be used to optimize the performance of artificial intelligence systems intended to imitate human responses. McKinstry gathered approximately 80,000 propositions that could be answered yes or no, e.g.: Is Earth a planet? Was Abraham Lincoln once President of the United States? Is the sun bigger than my foot? Do people sometimes lie? He called these propositions Mindpixels. These questions test both specific knowledge of aspects of culture, and basic facts about the meaning of various words and concepts. It could therefore be compared with the SAT, intelligence testing and other controversial measures of mental ability. McKinstry's aim was not to distinguish between shades of intelligence but to identify whether a computer program could be considered intelligent at all. According to McKinstry, a program able to do much better than chance on a large number of MIST questions would be judged to have some level of intelligence and understanding. For example, on a 20-question test, if a program were guessing the answers at random, it could be expected to score 10 correct on average. But the probability of a program scoring 20 out of 20 correct by guesswork is only one in 220, i.e. one in 1,048,576; so if a program were able to sustain this level of performance over several independent trials, with no prior access to the propositions, it should be considered intelligent. == Discussion == McKinstry criticized existing approaches to artificial intelligence such as chatterbots, saying that his questions could "kill" AI programs by quickly exposing their weaknesses. He contrasted his approach, a series of direct questions assessing an AI's capabilities, to the Turing test and Loebner Prize method of engaging an AI in undirected typed conversation. Critics of the MIST have noted that it would be easy to "kill" a McKinstry-style AI too, due to the impossibility of supplying it with correct answers to all possible yes/no questions by ways of a finite set of human-generated Mindpixels: the fact that an AI can answer the question "Is the sun bigger than my foot?" correctly does not mean that it can answer variations like "Is the sun bigger than (my hand | my liver | an egg yolk | Alpha Centauri A | ...)" correctly, too. However, the late McKinstry might have replied that a truly intelligent, knowledgeable entity (on a par with humans) would be able to work out answers such as (yes | yes | yes | don't know | ...) by applying its knowledge of the relative sizes of the objects named. In other words, the MIST was intended as a test of AI, not as a suggestion for implementing AI. It can also be argued that the MIST is a more objective test of intelligence than the Turing test, a subjective assessment that some might consider to be more a measure of the interrogator's gullibility than of the machine's intelligence. According to this argument, a human's judgment of a Turing test is vulnerable to the ELIZA effect, a tendency to mistake superficial signs of intelligence for the real thing, anthropomorphizing the program. The response, suggested by Alan Turing's essay Computing Machinery and Intelligence, is that if a program is a convincing imitation of an intelligent being, it is in fact intelligent. The dispute is thus over what it means for a program to have "real" intelligence, and by what signs it can be detected. A similar debate exists in the controversy over great ape language, in which nonhuman primates are said to have learned some aspects of sign languages but the significance of this learning is disputed.