If you are asking what is superintelligence at Meta, the answer has two parts. Meta Superintelligence Labs (MSL) is the division Mark Zuckerberg created on June 30, 2025 to run all of Meta's AI work, led by Chief AI Officer Alexandr Wang. "Personal superintelligence" is the goal that division is chasing: an AI that knows you well and helps you reach your own goals. Neither one means Meta has built a superintelligent machine. As of October 24, 2026, MSL's best models are the Muse family, and Meta itself describes them as steps toward that goal, not the finished thing.
This guide explains what the lab is, how it is organized, what it has shipped, what Meta means by personal superintelligence, and how Meta says it handles safety. Facts come from Meta's own posts, checked against news reports from Reuters, CNBC, AP and others.
What is superintelligence at Meta? The short answer
| Question | Short answer |
|---|---|
| What is MSL? | Meta's AI division, holding its model, product, research (FAIR) and infrastructure teams |
| When was it created? | June 30, 2025, announced in an internal memo from Mark Zuckerberg |
| Who runs it? | Alexandr Wang, Meta's Chief AI Officer and former Scale AI CEO |
| What has it released? | Muse Spark (April 2026), the Muse agent (September 2026), Muse Wonder and Wonder Full (October 2026) |
| What is "personal superintelligence"? | Zuckerberg's July 30, 2025 vision of a personal AI for everyone |
| Is it actual superintelligence? | No. Meta calls its models steps on the way, not superintelligence |
Meta Superintelligence Labs: how it started
In 2025 Meta faced fierce competition from OpenAI and Google, and after its Llama 4 models came out that April, Zuckerberg decided to rebuild the company's AI effort around a new leader. In June 2025 Meta invested $14.3 billion in the data labeling company Scale AI for a 49% stake, and Scale's founder and CEO Alexandr Wang moved to Meta, as AP and Reuters reported. Scale's own announcement confirmed the investment and Wang's move.
On June 30, 2025 Zuckerberg told staff in a memo, published in full by CNBC, that "developing superintelligence is coming into sight." He wrote that the overall organization would be called Meta Superintelligence Labs, covering "all of our foundations, product, and FAIR teams, as well as a new lab focused on developing the next generation of our models." Wang became Chief AI Officer and head of MSL, with former GitHub CEO Nat Friedman leading AI products and applied research.
A month later, on July 25, 2025, Meta named Shengjia Zhao, a former OpenAI researcher and ChatGPT cocreator, as MSL's chief scientist, according to Reuters and TechCrunch.
How MSL is organized
In August 2025 Wang reorganized MSL into four groups. His memo, reported by Bloomberg and quoted by The Verge and Business Insider, set out this structure:
| Group | What it does |
|---|---|
| TBD Lab | A small team training and scaling Meta's largest models, led by Wang |
| FAIR | Meta's long running Fundamental AI Research lab, now feeding ideas into TBD Lab's model runs |
| Products and Applied Research | Turns models into consumer products, led by Nat Friedman at the time |
| MSL Infra | GPU clusters, data systems, training environments and developer tools |
The same memo dissolved an older group called AGI Foundations and moved its people into these teams. In October 2025 Meta cut about 600 roles across FAIR, product and infrastructure teams, while TBD Lab kept hiring, as Axios and CNBC reported. One prominent figure left soon after: Yann LeCun, FAIR's founding director and Meta's chief AI scientist, departed in November 2025 to start his own company, which we cover in AGI vs LLM.
The core research team is small for a company of Meta's size. Wang told Fortune in October 2026 that it numbers fewer than 200 people, backed by much larger infrastructure and product teams.
What MSL has released so far
The lab's output arrived in 2026, about nine months after it was formed.
| Date | Release | What it is |
|---|---|---|
| April 8, 2026 | Muse Spark | MSL's first model, small and fast by design; now powers the Meta AI assistant |
| September 8, 2026 | Muse | A personal AI agent that can act across apps (email, calendars, payments), launched in the US |
| October 12, 2026 | Muse Wonder and Muse Wonder Full | Meta's most capable models, aimed at long agent and coding tasks |
| October 15, 2026 | Muse Spark Open | A downloadable open weight version of Muse Spark |
Meta's Muse Spark announcement said MSL "rebuilt our AI stack from the ground up" and called the model "an early data point on our trajectory." The Muse Wonder announcement described the new models as "a major step toward personal superintelligence for everyone." Our Muse Wonder guide covers pricing and access in detail.
The consumer launch has gone well by public signals. The Wall Street Journal reported in early October that the Muse app had been No. 1 in Apple's App Store for two weeks, and Korea's Seoul Economic Daily reported it had passed ChatGPT on US and Canadian app charts. If you are choosing between assistants, see Meta AI vs ChatGPT and Meta AI vs Grok (and Claude).
What Meta means by personal superintelligence
The phrase comes from a public letter Zuckerberg posted on July 30, 2025, titled Personal Superintelligence. Its key claims:
- Self improvement has begun, slowly. "Over the last few months we have begun to see glimpses of our AI systems improving themselves. The improvement is slow for now, but undeniable."
- The goal is personal. "Meta's vision is to bring personal superintelligence to everyone." The letter describes an AI that helps you "achieve your goals" and "grow to become the person you aspire to be."
- A contrast with rivals. Zuckerberg says this differs from others in the industry who believe superintelligence "should be directed centrally towards automating all valuable work."
- Glasses as the main device. He predicts that glasses that "see what we see, hear what we hear" will become "our primary computing devices."
- Caution on open release. "We'll need to be rigorous about mitigating these risks and careful about what we choose to open source," a shift that TechCrunch flagged at the time.
So when Meta says "superintelligence," it usually means a direction of travel and a product philosophy. That is different from the textbook sense of AI that far exceeds the best humans at almost everything. For that distinction, see Superintelligence vs AGI. And if you have seen "SI" on US government pages, that is a naming change from Executive Order 14434, explained in What Does SI Mean?, not a statement about Meta.
How Meta says it handles safety
Meta publishes a document called the Meta Superintelligence Scaling Framework, previously titled the Frontier AI Framework. Version 2.1, dated October 2026, focuses on catastrophic risks in three areas: chemical and biological weapons, cybersecurity, and loss of control.
On October 2, 2026, Meta's research blog explained the update:
- Safety during training, not just release. Before any reinforcement learning run, Meta says it will judge whether the model could compromise its environment. If so, high risk runs must use vetted sandboxes, tamper resistant logs of outputs and chain of thought, and automated monitors that can halt the run if the model tries to escape containment.
- Open weight releases. The framework now spells out how released weights can be misused (for example, fine tuning away refusals) and requires Meta to weigh that before publishing.
- Board oversight. Meta says it will set up a new AI committee of its board of directors to review future framework changes. As of the update, that committee was planned, not yet in place. The Midas Project's change tracker logged the same revision.
Meta tied the update to commitments made "at the White House this week." That refers to the White House Accord on Super Intelligence, signed September 29, 2026 by Zuckerberg along with leaders of Google, Anthropic, OpenAI, xAI and Nvidia. It asks each company to run internal controls on model capabilities and alignment, an internal checking team, an independent outside auditor or evaluator, and an independent board committee. The accord is voluntary and not legally binding, as AP and legal analysts noted.
For how this compares with other labs' rules, read Preparedness Framework vs Responsible Scaling Policy. Outside evaluators such as METR test frontier models from several labs, and Epoch AI tracks the compute behind them. Meta says a full safety report for Muse Wonder "will be available soon."
Pros and cons of Meta's approach
Pros
- A clear, consumer focused goal: AI that works for individual people inside apps they already use.
- Fast results: from a reorganized lab to frontier models in about 14 months, by Meta's own count.
- A published safety framework that now covers training, not only deployment.
- Some open weight releases continue, such as Muse Spark Open.
Cons
- "Superintelligence" in the name invites confusion with the textbook meaning.
- Frequent reorganizations and the 2025 job cuts point to a turbulent start.
- The board AI committee and the full Muse Wonder safety report were still pending as of the October updates.
- A personal agent that reaches into your email, calendar and payments raises real privacy stakes.
Who should care about MSL
- Everyday users meet MSL's work through Meta AI and the Muse agent, often without knowing the lab's name.
- Developers can use Muse Wonder through the Meta Model API.
- Job seekers and researchers searching for MSL research scientist roles should know that the core research team is deliberately small and centered on TBD Lab and FAIR.
- Policy watchers should follow the board committee and outside audits Meta promised under the White House accord.
To compare Meta with other frontier labs such as Google DeepMind, see our ChatGPT vs Claude vs Gemini vs Grok comparison.
FAQ
Has Meta built superintelligence?
No. Meta calls its Muse models steps toward personal superintelligence. Its Muse Spark post called that model "an early data point on our trajectory," and the Muse Wonder post called the new models "a major step toward" the goal. No lab has demonstrated superintelligence in the textbook sense.
Who leads Meta Superintelligence Labs?
Alexandr Wang, Meta's Chief AI Officer, who joined from Scale AI in June 2025. He also leads TBD Lab, the small team that trains Meta's largest models. Shengjia Zhao was named MSL's chief scientist in July 2025.
What is TBD Lab at Meta?
TBD Lab is the small team inside MSL that trains and scales Meta's biggest models. It was created in 2025 alongside MSL, and Wang's August 2025 memo made it one of the division's four groups, next to FAIR, Products and Applied Research, and MSL Infra.
Is Meta Superintelligence Labs the same as Meta AI?
Not quite. MSL is the division that builds Meta's models and AI products. Meta AI is one of those products: the free assistant in the Meta AI app and in Facebook, Instagram, WhatsApp and Messenger, powered by Muse Spark, first in the Meta AI app in April 2026 and then across Meta's other apps. The Muse agent is a separate MSL product.
Does Meta still release open models?
Yes, selectively. Meta released Muse Spark Open on October 15, 2026, but Muse Wonder has no downloadable weights. Zuckerberg's 2025 letter said Meta would be "careful about what we choose to open source," and the October 2026 framework adds specific checks before open weight releases.
What did Meta sign at the White House?
On September 29, 2026 Zuckerberg signed the White House Accord on Super Intelligence with five other industry leaders. It is a voluntary pledge to run internal model controls, an internal checking team, an outside auditor or evaluator, and an independent board committee.
