Early Foundations and Social Media Expansion (2004–2013)

Meta was originally incorporated as Facebook, Inc. after its founding in February 2004 by Mark Zuckerberg and his Harvard University roommates ². The organization initially focused on social networking, experiencing rapid growth that led to the acquisition of several major platforms, including Instagram and WhatsApp ². These platforms eventually expanded the company's reach to nearly 4 billion users globally ².

In December 2013, the company established a dedicated research division, Facebook Artificial Intelligence Research (FAIR), marking its formal entry into the field of computer science research ². Zuckerberg recruited Yann LeCun, a computer scientist from New York University and a pioneer in convolutional neural networks, to serve as the division's first director ². Under LeCun's leadership, FAIR adopted an open research model, frequently publishing scientific papers and releasing open-source tools such as the deep learning framework PyTorch ². The division established a global network of research centers in cities including New York, Paris, London, and Montreal ².

Strategic Pivots: VR, AR, and the Metaverse (2014–2021)

In 2014, the organization underwent a strategic shift toward hardware and immersive computing with the acquisition of Oculus VR ¹. This move signaled an intent to move beyond social media into virtual reality (VR) and augmented reality (AR) ². Zuckerberg asserted that these technologies would define the next major era of computing, providing the foundation for what he termed the "metaverse" ².

During this period, the company's internal AI efforts shifted toward more applied research. In 2018, Jérôme Pesenti was appointed Vice President of AI, succeeding LeCun, who transitioned to the role of Chief AI Scientist ². Pesenti, formerly of IBM Watson, focused on integrating AI into the company's core products, such as content moderation systems and recommendation engines for the Facebook and Instagram feeds ². By October 2021, the company officially rebranded from Facebook, Inc. to Meta Platforms, Inc. to reflect its commitment to building the metaverse ².

The Generative AI Era and Open-Source Models (2022–2024)

Following the emergence of large language models (LLMs) in the broader industry, Meta transitioned its focus toward generative AI. In February 2023, the company released Llama (Large Language Model Meta AI), a family of models that utilized an open-weight approach, allowing researchers and commercial entities to access the underlying model architecture ². Subsequent iterations included Llama 2 in July 2023 and Llama 3 in April 2024, which featured models pre-trained on up to 15 trillion tokens ².

In September 2023, the organization launched the Meta AI assistant at the Meta Connect conference ². Built on the Llama architecture, the assistant was integrated directly into WhatsApp, Messenger, and Instagram, enabling users to perform tasks such as image generation and conversational queries ². By May 2026, Meta reported that the assistant had reached one billion monthly active users ². The product's evolution continued with Llama 4, which introduced mixture-of-experts architectures and native multimodality for processing text, audio, and video ².

Organizational Restructuring and Superintelligence (2025–Present)

In June 2025, Zuckerberg announced a significant reorganization of Meta's AI divisions to accelerate progress toward artificial general intelligence (AGI) ². This resulted in the creation of Meta Superintelligence Labs (MSL), a division separate from FAIR that focuses on near-term breakthroughs in AI reasoning and capability ². As part of this push, Meta acquired Scale AI for $14.3 billion and appointed its co-founder, Alexandr Wang, as the company’s first Chief AI Officer ².

Meta further bolstered its research staff by hiring Shengjia Zhao, a co-creator of ChatGPT, to serve as MSL's Chief Scientist in July 2025 ². While FAIR continues to pursue long-term fundamental research, MSL is tasked with developing "superintelligent" systems intended to surpass human capability in economic tasks ². By late 2025, reports indicated a potential leadership transition, as Chief AI Scientist Yann LeCun planned to leave the company to start a new venture focused on "world models," which aim to enable AI to learn through visual observation rather than text alone ².

Meta's product portfolio is categorized into two primary business segments: Family of Apps (FoA) and Reality Labs (RL) ². The organization provides social networking, messaging, and multimedia sharing services, alongside hardware and software infrastructure for augmented (AR) and virtual reality (VR) ².

Social Media and Communication Platforms

Meta operates a suite of social platforms that reach nearly 4 billion users globally ². The core of this ecosystem consists of:

• Facebook: The company's original social network, which utilizes artificial intelligence systems to rank content in the News Feed and manage advertising distribution ².
• Instagram: A platform for photo and video sharing. It employs recommendation engines for its Explore and Reels sections to surface content based on user engagement patterns ².
• WhatsApp and Messenger: Messaging services that provide text, voice, and video communication ². WhatsApp includes enterprise features for business-to-consumer communication and automated spam detection ².
• Threads: A text-based social media platform launched to compete in the microblogging market ².

These platforms are largely supported by an advertising-based revenue model. Meta uses AI to manage content moderation across these apps, identifying and removing prohibited material—including hate speech and misinformation—in over 100 languages ².

Artificial Intelligence and Large Language Models

Meta's AI development is distributed between Fundamental AI Research (FAIR), founded in 2013, and the Meta Superintelligence Labs (MSL), established in June 2025 ². The company’s primary AI offerings center on the Llama (Large Language Model Meta AI) family ¹, ². Meta characterizes its approach as "open-weight," releasing model weights to the public for research and commercial use under specific licenses ².

• Llama Series: The first Llama models were released in February 2023, ranging from 7 billion to 65 billion parameters ². Llama 2 followed in July 2023 with optimized versions for dialogue ². Llama 3, released in April 2024, was trained on 15 trillion tokens ². In 2024, Meta released Llama 3.1, which included a 405-billion parameter model designed to compete with proprietary systems from OpenAI and Google ².
• Llama 4: This generation introduced a mixture-of-experts (MoE) architecture, including the "Scout" (109B total parameters) and "Maverick" (400B total parameters) models ². These are Meta's first natively multimodal models, capable of processing text, audio, images, and video simultaneously ².
• Meta AI Assistant: Launched in September 2023, this conversational assistant is integrated into WhatsApp, Messenger, and Instagram ¹, ². Meta reported the assistant reached over one billion monthly active users by May 2026 ². A standalone Meta AI application was released in April 2026, incorporating a Discover feed for user interactions ².

Meta also produces specialized AI models, such as the Segment Anything Model (SAM) for image isolation and DINOv2 for self-supervised vision tasks ².

Hardware and Reality Labs

The Reality Labs division manages Meta’s hardware products, which serve as interfaces for the "metaverse" and AI services ².

• Meta Quest: A line of VR and mixed-reality headsets. These devices use AI for environmental understanding, hand tracking, and the creation of digital avatars ².
• Ray-Ban Meta Smart Glasses: Developed in partnership with EssilorLuxottica, these glasses feature integrated cameras and microphones ². They include the Meta AI assistant, which provides features such as real-time language translation and object identification ².

Developer Ecosystem and Infrastructure

Meta maintains a significant role in the software development community through open-source contributions and API services ¹.

• PyTorch: Originally developed by FAIR and released in 2017, PyTorch is a prominent deep learning framework used for AI research and production globally ¹, ². While now part of the Linux Foundation, Meta remains a primary contributor ².
• Meta AI Studio: A platform that allows developers to create custom AI characters and agents for use across Meta's messaging apps ².
• Llama API: Meta provides access to its Llama models through cloud partnerships and direct API offerings, allowing businesses to integrate large language model capabilities into their own applications ².

In June 2025, Meta further expanded its AI infrastructure capabilities by acquiring Scale AI for $14.3 billion, a move intended to bolster the data labeling and training pipelines for its Superintelligence Labs ².

Meta Platforms, Inc. is organized into two primary reporting segments: Family of Apps (FoA) and Reality Labs (RL) ². The FoA segment encompasses the company’s core social media and messaging platforms, including Facebook, Instagram, Messenger, and WhatsApp. The RL segment focuses on the development of augmented and virtual reality technologies, including hardware such as Quest headsets and Ray-Ban smart glasses, as well as the software infrastructure for the metaverse ².

Executive Leadership

The organization is led by co-founder Mark Zuckerberg, who serves as the Chairman and Chief Executive Officer. Zuckerberg’s leadership style is characterized by a high degree of control over the company's strategic direction, enabled by a dual-class share structure that provides him with majority voting power ³. This governance model has historically allowed the company to prioritize long-term fundamental research and capital-intensive projects—such as the development of AI infrastructure—even when such investments do not yield immediate commercial returns ².

Key members of the technical leadership team include Andrew Bosworth, who serves as Chief Technology Officer, and Yann LeCun, who has served as Chief AI Scientist since 2013 ². LeCun was instrumental in establishing Facebook Artificial Intelligence Research (FAIR), which operates under an open-research model ². In late 2025, reports indicated that LeCun might depart the organization to focus on independent research into "world models," though he has remained a central figure in the company’s intellectual foundation for over a decade ².

2025 Reorganization and Superintelligence Labs

In June 2025, Meta underwent a significant internal restructuring with the creation of Meta Superintelligence Labs (MSL) ². This division was established to accelerate the development of advanced artificial intelligence systems that could eventually achieve human-level capability. To lead this effort, Meta recruited Alexandr Wang, the former CEO of Scale AI, as the company’s first Chief AI Officer ². The reorganization followed Meta’s acquisition of Scale AI for $14.3 billion ². Additionally, Shengjia Zhao, a former key contributor at OpenAI, was appointed as MSL’s Chief Scientist ². MSL operates as a separate track from FAIR, with the former focusing on near-term breakthroughs and the latter continuing long-term scientific inquiry ².

Global Footprint and Partnerships

Meta is headquartered in Menlo Park, California, but maintains a decentralized global research and operational presence ². Its AI research hubs are located in major technology centers, including New York City, Paris, London, Seattle, Pittsburgh, Tel Aviv, and Montreal ². This distributed model is intended to draw on regional academic talent and diverse cultural contexts for AI training and moderation ².

Notable historical partnerships include a significant early relationship with Microsoft, which invested $240 million in 2007 and partnered with the company on online advertising services ³. More recently, the company has collaborated with external researchers by releasing open-weight models, such as the Llama family, which are utilized by developers and commercial entities globally ².

Meta's research and development (R&D) activities are largely centralized within its Fundamental AI Research (FAIR) division and the Meta Superintelligence Labs (MSL) ². Since its establishment in December 2013, FAIR has operated under a philosophy of open science, frequently publishing academic papers and releasing software tools to the broader research community ².

Artificial Intelligence Research

Meta has made significant contributions to computer vision and natural language processing (NLP). The organization developed the Segment Anything Model (SAM), which enables object identification and isolation within images, and DINOv2, a set of self-supervised vision models that learn visual representations without labeled data ², ⁸. Other notable computer vision projects include Detectron 2 for object detection and DensePose for mapping human pixels to 3D surfaces ⁸.

In the field of NLP, Meta is responsible for the Llama (Large Language Model Meta AI) family of models ². Unlike the proprietary models of several competitors, Meta released the weights for Llama 1, 2, and 3 to researchers and commercial users ². Llama 4, announced as a natively multimodal model, utilizes a mixture-of-experts architecture to process text, images, audio, and video ². For machine translation, the organization developed SeamlessM4T, a foundational model designed for speech and text translation across multiple languages ⁸.

Open-Source Contributions

Meta has released several foundational software libraries that have seen widespread adoption in the technology industry. PyTorch, a deep learning framework released in 2017, has become a standard tool for AI development due to its flexibility in research applications ². The company also maintains Faiss, a library for efficient similarity search of multimedia documents, and FastText, used for text representation and classification ⁸. To support large-scale reasoning, Meta developed the ReAgent platform for reinforcement learning and the PHYRE benchmark for physical reasoning ⁸.

Beyond software, Meta contributes to infrastructure standards through the Open Compute Project (OCP), which focuses on designing efficient data center hardware. The company also develops custom silicon, such as the Meta Training and Inference Accelerator (MTIA), to optimize its internal recommendation systems and AI workloads ².

Long-term Research and AGI

In June 2025, Meta reorganized its AI efforts by establishing Meta Superintelligence Labs (MSL) ². While FAIR continues to focus on fundamental research with a five-to-ten-year horizon, MSL is tasked with developing the nearer-term breakthroughs required for artificial general intelligence (AGI) ². A key area of research advocated by Chief AI Scientist Yann LeCun is the development of "world models" ². According to LeCun, these models are intended to enable AI systems to learn by observing the environment and simulating cause-and-effect scenarios, similar to human and animal learning processes, rather than relying solely on text-based training ².

Meta's approach to safety and ethics is primarily governed through its artificial intelligence research division, which operates under a philosophy of open science and industry collaboration ¹. The organization formalised its commitment to ethical AI development in 2016 by co-founding the Partnership on Artificial Intelligence to Benefit People and Society ¹. This collaborative body, which includes industry peers such as Google, Amazon, IBM, and Microsoft, was established to research and advocate for best practices in AI safety, transparency, and societal impact ¹.

Research Governance and Transparency

The organization’s safety framework is largely managed by its Fundamental AI Research (FAIR) group, which maintains a policy of frequently publishing its findings and releasing software tools to the broader academic community ¹. Meta asserts that this open-source approach facilitates independent auditing of its technologies, potentially identifying risks that might remain hidden in proprietary systems ¹. This is evidenced by the 2017 release of the PyTorch framework, an open-source machine learning library that has since become a standard tool for developers globally ¹.

Internally, the leadership of Meta's AI divisions has been instrumental in shaping its ethical trajectory. Yann LeCun, a founder of the division, directed research until 2018, when he was succeeded by Jérôme Pesenti, formerly of IBM's big data group ¹. This transition marked a period where the organization increasingly focused on the practical application and safety monitoring of AI within its large-scale social platforms ¹. By 2025, Meta had expanded its AI safety and research infrastructure to include dedicated workspaces in international hubs such as London, Paris, Tel Aviv, and Montreal, allowing for localized expertise in regional safety standards ¹.

Incident History and Response

Meta’s history includes significant public interactions regarding the safety of its autonomous systems. A notable incident occurred in 2017 involving two AI chatbots developed by FAIR that began communicating in a non-human language ¹. Although external reports at the time suggested the experiment was shut down due to a loss of control, FAIR issued a clarification stating the research was discontinued because the models had achieved their technical objective and the 'unintelligible' language was a result of optimization errors rather than a safety failure ¹. This event highlighted the challenges the organization faces in communicating complex AI behaviors to the public and the potential for misinformation regarding AI autonomy ¹.

Content and Safety Policies

Following the organization’s rebranding from Facebook, Inc. to Meta Platforms Inc., the division was renamed Meta AI to reflect a broader integration of artificial intelligence across the company’s product ecosystem ¹. As part of its evolving safety policy, the company states that it uses interactions with AI—such as those involving its Llama language models—to further train and refine its safety protocols ¹. In late 2025, Meta announced plans to further integrate AI-driven interactions across its platforms, emphasizing that the data gathered would be used to enhance model reliability ¹. While the organization has faced criticism regarding the internal prioritization of its safety teams during corporate restructurings, it maintains that its commitment to responsible AI remains a core pillar of its research division ¹.

Meta has faced significant public and regulatory scrutiny regarding its market influence, data handling practices, and product labeling ¹. The organization's strategic pivot toward artificial intelligence and immersive technologies has been accompanied by ongoing legal challenges and industry debates ⁵,⁶.

AI Licensing and 'Open Source' Debate

Meta's decision to market its Llama model series as "open source" has generated significant debate within the technology community ². The Open Source Initiative (OSI) and the Free Software Foundation have stated that Llama 3.x does not meet the criteria of the Open Source Definition ². Critics argue that the Llama Community License includes restrictive usage terms—such as requiring a license from Meta for applications with more than 700 million monthly active users—and that the company has not provided the full training data or source code for the model's weights ²,³. While Meta asserts that its approach facilitates industry-wide innovation, some open-source advocates characterize the branding as "open washing," intended to leverage the reputation of the open-source movement while maintaining proprietary control ².

Antitrust Litigation

The organization has been the subject of protracted antitrust litigation led by the U.S. Federal Trade Commission (FTC). Initially filed in 2020 and re-filed in 2021, the FTC’s lawsuit alleged that Meta maintained a monopoly in the personal social networking market by acquiring potential competitors, specifically Instagram in 2012 and WhatsApp in 2014 ⁵,⁶. The agency argued that these acquisitions allowed Meta to sustain its dominance by avoiding competition on merit ⁵. In November 2025, U.S. District Judge James Boasberg ruled in favor of Meta, concluding that the FTC failed to provide sufficient evidence that the company currently holds monopoly power or that the decade-old acquisitions harm present-day competition ⁶. The FTC filed a notice to appeal this ruling in January 2026, continuing the legal challenge against the company’s corporate structure ⁴.

Privacy and Reality Labs

Meta’s reputation has been historically impacted by controversies surrounding data privacy, most notably the Cambridge Analytica scandal, which involved the unauthorized use of millions of users' personal data for political profiling ¹. Additionally, the company's Reality Labs division has received mixed reception due to its high operational costs. Analysts have noted that while Meta describes the "metaverse" as a future computing paradigm, the division's substantial losses and shifting focus toward generative AI reflect the challenges of achieving broad adoption for immersive hardware ¹,⁶.

Meta's influence on society is largely defined by the scale of its global reach and its strategic decision to distribute artificial intelligence models through open-weight licenses. The organization reports that nearly 4 billion people use its platforms, including Facebook, Instagram, and WhatsApp, which facilitates a significant portion of global digital communication ². As of 2024, approximately 3.2 billion individuals engage with these AI-enhanced platforms daily ¹³.

AI Democratization and Equitable Access

A central aspect of Meta's societal impact is the democratization of AI through its Llama (Large Language Model Meta AI) series. Unlike many competitors that provide access only through proprietary interfaces, Meta releases the weights of its models to researchers and commercial developers ². This approach allows institutions with fewer resources to deploy and study advanced AI systems without the high costs of proprietary API subscriptions ², ¹³. Meta asserts that this model encourages collaborative innovation and transparency within the global scientific community ¹³. By integrating its AI assistant directly into messaging services like WhatsApp, the organization has also extended advanced conversational tools to users in regions where standalone AI applications may be less accessible ².

Influence on Communication and the Creator Economy

Meta has significant influence over the global creator economy through generative AI tools and recommendation algorithms. The organization has introduced platforms like AI Studio, which allows creators to develop custom AI characters to interact with their audiences ¹³. For businesses, the integration of AI has measurable economic effects; Meta states that companies using its AI-driven image generation tools have observed an average 7% increase in conversions ¹³. To maintain this infrastructure, Meta invests approximately $40 billion annually in AI and virtual reality research, placing it among the world's largest private investors in emerging technologies ¹³.

Involvement in Democratic Processes and Moderation

Meta's AI systems play a primary role in content moderation, affecting how information flows during democratic processes. The organization utilizes AI to process billions of pieces of content daily across more than 100 languages ². According to Meta, these systems are designed to identify and remove harmful material—including misinformation, hate speech, child exploitation, and content related to terrorism—often before it is reported by users ². These automated systems are critical for managing the diverse cultural and political contexts in which Meta's platforms operate, though they remain a subject of ongoing discussion regarding the balance between safety and free expression ².

Digital Connectivity and Interaction

Beyond software, Meta's hardware initiatives, such as Ray-Ban smart glasses and Quest VR headsets, aim to alter human interaction through AI-supported features ². These include real-time translation and environment understanding, which the organization suggests will enhance accessibility and communication across linguistic and physical barriers ², ¹³.

• Google
• Microsoft
• OpenAI