Meta Superintelligence Labs

Meta Superintelligence Labs (MSL) is the artificial intelligence division that Meta created on June 30, 2025 to consolidate its research, foundation models, and AI products under a single organization aimed at building what the company calls personal superintelligence ^[1][2]. Mark Zuckerberg announced MSL in an internal memo and named Alexandr Wang, the former chief executive of the data labeling company Scale AI, as Meta's first Chief AI Officer and head of the new group, with the former GitHub chief executive Nat Friedman co-leading its AI products and applied research work ^[1][3]. The reorganization folded Meta's longstanding research lab, Facebook AI Research (FAIR), its Llama and generative AI product teams, and a new frontier model unit into one structure, and it was paired with a roughly 14.3 billion dollar investment that gave Meta about a 49 percent stake in Scale AI at a valuation above 29 billion dollars ^[2][4][5].

In the memo Zuckerberg wrote that "as the pace of AI progress accelerates, developing superintelligence is coming into sight," and he called Wang "the most impressive founder of his generation" ^[1]. The move kicked off one of the most aggressive talent recruitment drives the field has seen, with reported compensation offers that drew public commentary from rival labs, and it was followed within months by a contraction: in October 2025 Meta cut about 600 roles in the unit, leaving MSL with just under 3,000 employees ^[6][7][9]. By late 2025 the organization had reshaped its reporting lines and lost or risked losing several prominent researchers, including Yann LeCun, the Turing Award winner who founded Meta's original research lab ^[8][9]. In 2026 the lab shipped the Muse family of models, beginning with Muse Spark in April, and on September 8, 2026 it launched Muse, a consumer personal AI agent powered by Muse Spark 1.3 ^[19][31][49].

What did Meta's AI effort look like before MSL?

Meta had invested in artificial intelligence for more than a decade before MSL existed. FAIR was created in 2013 under LeCun, and it became one of the most cited academic style labs in the industry. Separately, Meta built large product and applied teams that shipped the Meta AI assistant across Facebook, Instagram, and WhatsApp, and that produced the open weight Llama family of language models. Meta had positioned Llama as a counterweight to closed systems from OpenAI, Google DeepMind, and Anthropic, and Zuckerberg had framed open releases as a strategic advantage.

That strategy ran into trouble in early 2025. The Llama 4 release in April 2025 received a muted reception, and the largest model in the family, code named Behemoth, was delayed amid reports that its performance fell short of internal targets ^[10][29]. Zuckerberg reportedly grew frustrated with the pace of progress and the structure of the existing teams. Press accounts describe him personally driving a recruitment and reorganization effort over the spring of 2025, holding meetings with prospective hires and rethinking how Meta's AI work was organized ^[6][11]. MSL was the result of that effort.

How was MSL formed, and what was the Scale AI deal?

The public starting point for MSL was Meta's investment in Scale AI. In June 2025 Meta agreed to pay about 14.3 billion dollars for a stake of roughly 49 percent in Scale AI, a deal that valued the startup at more than 29 billion dollars ^[4][5]. The structure was a non controlling minority investment rather than an outright acquisition, an arrangement that several reports noted helped limit the antitrust scrutiny a full takeover might have attracted ^[4][27]. As part of the deal, Wang stepped down as Scale AI's chief executive to join Meta, while remaining on the Scale AI board, and Jason Droege, Scale AI's chief strategy officer, became interim chief executive of the startup ^[5].

A few weeks later, on June 30, 2025, Zuckerberg sent the memo that formally created Meta Superintelligence Labs and named Wang as Chief AI Officer ^[1][2]. In the memo Zuckerberg described Wang as "the most impressive founder of his generation" and said Nat Friedman would partner with Wang to lead Meta's work on AI products and applied research ^[1][3]. Zuckerberg framed the goal as building personal superintelligence for everyone and argued that Meta was well placed to pursue it given its compute resources, its distribution across billions of users, and its experience shipping AI products at scale ^[1][2].

Who leads Meta Superintelligence Labs, and how is it organized?

MSL gathered Meta's AI work under a small group of leaders. Wang held overall responsibility as Chief AI Officer. Friedman led product and applied research. Daniel Gross, who had been chief executive of the startup Safe Superintelligence co founded by Ilya Sutskever, joined alongside Friedman, with whom he had run a venture investment firm called NFDG ^[12]. On July 25, 2025, Wang announced that Shengjia Zhao, a former OpenAI researcher who had contributed to ChatGPT and to OpenAI's reasoning models, would serve as Chief Scientist of MSL ^[13]. That created a structure in which Zhao led the science of the frontier model work while LeCun continued to hold the Chief AI Scientist title associated with FAIR.

The organization was built around four units. A new group called TBD Lab, led by Wang, took on frontier and next generation model development, including future Llama models. FAIR remained the longer horizon research arm. A products and applied research group, associated with Friedman, focused on the Meta AI assistant and related applications. An infrastructure group, led by Aparna Ramani, handled the data centers, compute, and systems needed to train and serve large models ^[9][11].

The table below summarizes the leadership reported across 2025.

The table below sketches the four reported units.

Several of these roles changed in 2026. On January 12, 2026, Zuckerberg created Meta Compute, a top-level organization for Meta's data center buildout, co-led by head of global infrastructure Santosh Janardhan and Daniel Gross; Gross was given a new group responsible for long-range capacity planning and the supply chain for chips, servers, and networking gear ^[44]. In March 2026 the team behind Dreamer, an agent-building startup co-founded by David Singleton and Hugo Barra, joined MSL under an arrangement in which Dreamer licensed its technology to Meta rather than being acquired outright ^[47]. Singleton later spoke for MSL in public about the security of the Muse agent ^[41]. On June 25, 2026, Dawn Song, a computer science professor at the University of California, Berkeley and co-founder of the AI security startup Virtue AI, announced on X that she would join MSL as Vice President of AI Research "together with many members of the Virtue AI team" to help shape Meta's AI safety and AI security efforts ^[51][52]. On August 24, 2026, Axios reported that Luke Metz, a researcher who had left OpenAI for Thinking Machines in 2024 and rejoined OpenAI earlier in 2026, had joined MSL and would report to Wang ^[48].

How big were the compensation offers in Meta's talent war?

MSL was staffed in part through an unusually public recruiting push. In his June 30 memo, Zuckerberg named a group of about eleven new hires drawn from OpenAI, Google DeepMind, Anthropic, and other leading labs ^[1][2]. The named recruits included Trapit Bansal, Shuchao Bi, Huiwen Chang, Ji Lin, Hongyu Ren, and Jiahui Yu from OpenAI, Jack Rae and Pei Sun from Google DeepMind, Joel Pobar from Anthropic, and Johan Schalkwyk from the voice startup Sesame AI ^[1][2]. Separately, Ruoming Pang, who had led Apple's foundation models team, was reported to have joined Meta ^[14].

The compensation attached to these moves became a story in itself. Sam Altman, the chief executive of OpenAI, said on the "Uncapped" podcast that Meta had offered signing bonuses as large as 100 million dollars to some OpenAI staff, along with larger annual pay, adding that "so far none of our best people have decided to take them up on that" ^[6][7]. Subsequent reports described multiyear packages for a few senior researchers that ran into the tens or hundreds of millions of dollars ^[6][14]. Meta executives disputed parts of this account, with some saying the 100 million dollar figure was not an accurate description of how the offers were structured ^[6][7]. These compensation figures come from press reporting and from comments by interested parties, and the exact terms of individual deals have not been disclosed by Meta.

The recruiting drive was tied to a large increase in spending on computing. Meta narrowed its 2025 capital expenditure guidance to a range of 66 billion to 72 billion dollars, raising the low end of the range, with much of the spending going to AI infrastructure ^[15]. Zuckerberg said Meta was building multi gigawatt data center clusters, naming one called Prometheus expected to come online in 2026 and a larger one called Hyperion that he said could scale up to five gigawatts over several years ^[26].

What is the personal superintelligence vision?

In late July 2025 Zuckerberg published a short public essay titled Personal Superintelligence that set out the thinking behind MSL ^[16]. He argued that Meta's aim differed from rivals who framed advanced AI mainly as a way to automate economically valuable work. "This is distinct from others in the industry who believe superintelligence should be directed centrally towards automating all valuable work, and then humanity will live on a dole of its output," he wrote ^[16]. Meta's focus, he argued, would be a personal superintelligence that knows people deeply, understands their goals, and helps them pursue those goals, delivered through personal devices such as glasses and assistants rather than as a centralized service ^[16]. The framing connected the AI work to Meta's existing hardware ambitions and to its large consumer reach.

The essay also signaled a shift on open source. Meta had long released Llama weights openly, and Zuckerberg had defended that approach as both strategic and good for the field. In the Personal Superintelligence essay he struck a more cautious note, suggesting Meta would be more selective about what it open sources as systems grow more capable, citing safety concerns ^[16]. Observers read this as a notable change for a company that had built much of its AI identity around open releases ^[17]. MSL's first frontier models, the Muse Spark series, were offered only as hosted services. On August 10, 2026, however, the lab released Muse Glimmer, a 30-billion-parameter model trained on Muse Spark's outputs and aimed at local agent workloads, with open weights under the Apache 2.0 license ^[43]. Meta's September 2, 2026 post announcing Muse Spark 1.3 listed "the Muse Spark open weights release" on its roadmap, without giving a date ^[33].

How was MSL received, and who left?

Reaction to MSL was mixed. Supporters pointed to the concentration of talent, capital, and compute that few competitors could match, and to Meta's record of shipping AI features to a very large user base. Critics questioned whether assembling a roster of highly paid stars would translate into research breakthroughs, arguing that culture and cohesion matter and that money alone does not guarantee results; within two months of the unit's launch, several researchers had resigned, two of them returning to OpenAI ^[28]. Some pointed to the demands the push placed on workers and the scale of the spending, and others highlighted the apparent move away from open source as a risk to the developer community that had adopted Llama ^[17]. A broader strain of skepticism focused on the pursuit of superintelligence itself, with critics questioning whether it is achievable or even well defined ^[17].

The reorganization also produced friction inside Meta. LeCun, who had founded FAIR and who has long argued that large language models alone will not reach human level intelligence, was reported to now report to Wang under the new structure, where he had previously reported to chief product officer Chris Cox ^[30]. On November 19, 2025, LeCun confirmed he would leave Meta after about twelve years to start his own company focused on world models, the research direction he had championed ^[24]; that venture, Advanced Machine Intelligence (AMI) Labs, was set up with Alex LeBrun as chief executive and LeCun as executive chair ^[25]. Joelle Pineau, a vice president who had led FAIR, had announced her departure earlier in 2025, before the reorganization was complete ^[9][18]. Several researchers connected to the Llama and FAIR teams left during this period ^[9][28]. Departures continued in 2026. Reuters, citing The Information, reported on February 25, 2026 that Ruoming Pang, who had overseen AI infrastructure for MSL after joining from Apple about seven months earlier, had left Meta for OpenAI ^[45]. On September 9, 2026, Semafor, citing a person briefed on the matter, reported that Andrew Tulloch, a researcher who had come to Meta from Thinking Machines Lab and worked in TBD Lab, was leaving the company after delaying his departure until the Muse launch ^[46].

MSL went through its own contraction not long after its expansion. On October 22, 2025 Meta cut about 600 roles within the AI organization as part of an effort to make it leaner, leaving the unit with just under 3,000 employees ^[7][9]. Reporting indicated the cuts fell more heavily on FAIR and on product and infrastructure teams, while the newer TBD Lab was largely spared ^[7][9][11]. In an internal memo, Wang wrote that "by reducing the size of our team, fewer conversations will be required to make a decision, and each person will be more load-bearing and have more scope and impact" ^[23]. The contrast between cutting hundreds of existing roles and paying premium packages for new hires drew comment ^[9][17].

What has MSL shipped?

MSL shipped its first frontier model, Muse Spark, on April 8, 2026, a proprietary multimodal reasoning model that was also Meta's first flagship release without open weights ^[19]. Muse Spark 1.1 followed on July 9, 2026, together with a public developer preview of the Meta Model API ^[20]. On August 5, 2026 the lab released Muse Code, a terminal coding agent in beta and Meta's first coding agent, powered by the new coding-focused Muse Spark 1.2 model ^[21][22]. Wang said, per CNBC's reporting, that Meta was positioning the tool and model family to compete on price with Anthropic's Claude Code and OpenAI's Codex rather than on bleeding-edge capability ^[22].

The pace increased in late summer 2026. The open-weight Muse Glimmer followed on August 10 ^[43]. On September 2, 2026, MSL released Muse Spark 1.3, which it described as an update for longer-horizon agentic and coding work; in comparisons by Meta engineers, it used about 20 percent fewer tool calls and about 25 percent fewer tokens than Muse Spark 1.2 ^[33]. The model's max reasoning mode, which Meta initially held back for additional safety testing, was released publicly on September 4 ^[34][50].

Muse personal agent

On September 8, 2026, Meta launched Muse, which it called a personal AI agent that can carry out tasks rather than only answer questions, powered by Muse Spark 1.3 ^[31][49]. Meta said Muse can open a browser, fill out forms, send email, and make purchases, keeps working after the user closes the app, and asks for approval before sensitive actions such as sending an email or completing a purchase. Each user's agent runs in a dedicated cloud virtual machine that Meta calls Muse Secure VM, with a separate Sentinel agent that must approve anything Muse sends to the internet ^[31]. At launch it was rolling out in the United States on iOS, Android, and the muse.ai website, and could also be used inside WhatsApp; Meta said it was "free for most of what people need," with paid subscription plans for heavier use ^[31]. MSL's technical post on Muse's safety design said the lab had been building and using the agent internally since early 2026 and that "Hatch is our internal name for Muse in the codebase." The same post opened a public bug bounty paying up to $300,000 for valid reports ^[32].

The rollout widened quickly. Zuckerberg announced a macOS app on September 17 ^[37], and on September 18 the official Muse account said the agent was available in Canada on iOS and the web, with Android "coming soon" ^[35][36]. Adoption figures came from third-party analytics firms, not from Meta. According to Sensor Tower, as reported by CNBC, Muse overtook ChatGPT as the top free iOS app in the United States on September 18 and had passed 2.5 million downloads by September 21 ^[38]. Apptopia, as reported by TechCrunch, estimated that Muse was downloaded 1.8 million times on iOS in the United States and Canada in its first 12 days, against 1.3 million for ChatGPT in the same period after its own iOS launch; TechCrunch noted that Apptopia has no access to Meta's internal figures and that Meta had not published adoption numbers ^[39].

Commerce was a central part of the pitch. At launch Muse paid for purchases with Stripe's Link, and Meta listed Shop Pay as coming soon ^[31]. Amazon blocked Muse from its shopping site, saying it had not been told the agent would access its store and that Muse appeared to capture and store customer credentials ^[38]. Shopify announced an agentic checkout partnership with Muse on September 21 ^[38], and PayPal, Expedia, and Instacart followed; early on September 23 (UTC) the official Muse account listed all four as new connectors, "all coming soon" ^[40].

Security and privacy problems surfaced in the second week. On September 22, The Verge, following a report by Ars Technica, said security researcher Patrick Wardle had found a zero-day flaw in the Muse Mac app that let code already running on a user's computer redirect Muse's transcription to an attacker's server and gain access to the Muse account. Meta issued a hotfix, and Singleton called it "a local privilege escalation attack, not a remote exploit" with low practical risk ^[41]. The same day Reuters reported, from internal posts, that Meta had been testing a "human concierge" in which contractors placed some phone calls requested through Muse; a vice president in MSL acknowledged that starting the test without proper disclosures "was a miss" and said the feature had been rolled back for now ^[42].

Why does MSL matter?

Meta Superintelligence Labs marked a clear shift in how Meta organizes and talks about its AI work. By putting FAIR, the Llama and product teams, and a dedicated frontier unit under a single Chief AI Officer recruited from outside, the company centralized decisions that had previously sat across separate groups, and it tied that structure to a sharply higher level of spending on talent and compute. The branding around superintelligence and the personal superintelligence framing also repositioned Meta's public message, moving it closer to the language used by frontier labs while keeping a distinct emphasis on consumer devices and reach.

Whether the bet would pay off was still an open question in early 2026. By September 2026 there were early commercial signals: third-party estimates put Muse downloads above 2.5 million in its first two weeks ^[38], and Reuters reported on September 22 that Meta's shares had risen more than 20 percent since the Muse launch, adding more than $200 billion in market value ^[42]. Those figures measure early interest and investor reaction rather than long-term use, and the same weeks brought the Mac app vulnerability and the human concierge test described above ^[41][42]. The early months brought visible turbulence, including the contraction of legacy research teams and the reported departures of senior figures associated with Meta's original AI identity. The first concrete tests arrived with the Muse Spark models in 2026, and the open question is whether the assembled talent can deliver systems competitive with those from OpenAI, Google DeepMind, and Anthropic. For Meta, MSL represents a large and public wager that consolidating leadership, money, and compute can reset an AI effort that had fallen behind some of its rivals.

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Heath, Alex. "Read Mark Zuckerberg's memo announcing Meta Superintelligence Labs." CNBC, June 30, 2025. cnbc.com/...a-superintelligence-labs-read-the-memo
2. ^1 ^2 ^3 ^4 ^5 ^6"Meta deepens AI push with 'Superintelligence' lab, source says." Reuters (Yahoo Finance syndication), June 30, 2025. finance.yahoo.com/...h-superintelligence-194647786
3. ^1 ^2Kylie Robison. "Here Is Everyone Mark Zuckerberg Has Hired So Far for Meta's 'Superintelligence' Team." Wired, June 30, 2025. wired.com/...rberg-welcomes-superintelligence-team
4. ^1 ^2 ^3Ivan Mehta. "Scale AI confirms 'significant' investment from Meta, says CEO Alexandr Wang is leaving." TechCrunch, June 13, 2025. techcrunch.com/...ays-ceo-alexandr-wang-is-leaving
5. ^1 ^2 ^3"Founder, Alexandr Wang, Joins Meta to Work on AI Efforts." Scale AI, June 12, 2025. scale.com/...unces-next-phase-of-company-evolution
6. ^1 ^2 ^3 ^4 ^5Isaac, Mike. "Mark Zuckerberg Ramps Up Meta's A.I. Spending as Competition Heats Up." The New York Times, June 27, 2025. nytimes.com/...mark-zuckerberg-meta-ai
7. ^1 ^2 ^3 ^4 ^5"Sam Altman says Meta offered OpenAI staff $100 million bonuses, as Mark Zuckerberg ramps up AI poaching efforts." CNBC, June 18, 2025. cnbc.com/...th-100-million-bonuses-mark-zuckerberg
8. ^"Meta's chief AI scientist Yann LeCun reportedly plans to leave to build his own startup." TechCrunch, November 11, 2025. techcrunch.com/...o-leave-to-build-his-own-startup
9. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8"Meta lays off 600 from 'bloated' AI unit as Wang cements leadership." CNBC, October 22, 2025. cnbc.com/...meta-layoffs-ai
10. ^Kyle Wiggers. "Meta releases Llama 4, a new crop of flagship AI models." TechCrunch, April 5, 2025. techcrunch.com/...a-new-crop-of-flagship-ai-models
11. ^1 ^2 ^3Riley Griffin. "Meta Restructures AI Group Again in Pursuit of Superintelligence." Bloomberg, August 19, 2025. bloomberg.com/...n-in-pursuit-of-superintelligence
12. ^"Meta reportedly plans big new AI hires." Axios, June 20, 2025. axios.com/...meta-zucker-berg-ai-daniel-gross
13. ^Maxwell Zeff. "Meta names Shengjia Zhao as chief scientist of AI superintelligence unit." TechCrunch, July 25, 2025. techcrunch.com/...ist-of-ai-superintelligence-unit
14. ^1 ^2"Apple Loses Top AI Models Executive to Meta's Hiring Spree." Bloomberg, July 7, 2025. bloomberg.com/...-executive-to-meta-s-hiring-spree
15. ^"Meta Reports Second Quarter 2025 Results." Meta Platforms press release (PR Newswire), July 30, 2025. prnewswire.com/...d-quarter-2025-results-302517817
16. ^1 ^2 ^3 ^4Zuckerberg, Mark. "Personal Superintelligence." Meta, July 30, 2025. about.fb.com/...personal-superintelligence
17. ^1 ^2 ^3 ^4Matthew Urwin. "Meta Superintelligence Labs: What We Know So Far." Built In, August 20, 2025 (updated April 2026). builtin.com/...meta-superintelligence-labs
18. ^Jonathan Vanian. "Meta's head of AI research announces departure." CNBC, April 1, 2025. cnbc.com/...ead-of-ai-research-announces-departure
19. ^1 ^2 ^3"Introducing Muse Spark: Scaling Towards Personal Superintelligence." Meta Superintelligence Labs, April 8, 2026. ai.meta.com/...introducing-muse-spark-msl
20. ^1 ^2"Introducing Muse Spark 1.1 and the Meta Model API." Meta Superintelligence Labs, July 9, 2026. ai.meta.com/...introducing-muse-spark-meta-model-api
21. ^1 ^2"Introducing Muse Code and Muse Spark 1.2." Meta Superintelligence Labs, August 5, 2026. research.meta.ai/...g-muse-code-and-muse-spark-1-2
22. ^1 ^2Vanian, Jonathan. "Meta debuts first AI coding agent to take on Anthropic and OpenAI." CNBC, August 5, 2026. cnbc.com/...-code-to-take-on-anthropic-and-openai-
23. ^"Exclusive: Meta slashes jobs in its older AI operations." Axios, October 22, 2025. axios.com/...meta-superintelligence-tbd-ai-reorg
24. ^1 ^2Vanian, Jonathan. "Meta chief AI scientist Yann LeCun is leaving to create his own startup." CNBC, November 19, 2025. cnbc.com/...ist-yann-lecun-is-leaving-the-company-
25. ^Schmelzer, Ron. "Yann LeCun's New Startup AMI Labs: Can World Models Move Beyond Hype?" Forbes, December 22, 2025. forbes.com/...bs-can-world-models-move-beyond-hype
26. ^Capoot, Ashley. "Meta CEO Zuckerberg says first AI data supercluster will come online in 2026." CNBC, July 14, 2025. cnbc.com/...meta-zuckerberg-ai
27. ^"Meta's big AI deal could invite antitrust scrutiny." Axios, June 11, 2025. axios.com/...as-noncontrol-scale-ai-deal-antitrust
28. ^1 ^2Yildirim, Ece. "Meta's Already Bleeding AI Talent Two Months Into Hiring Spree." Gizmodo, August 27, 2025. gizmodo.com/...months-into-hiring-spree-2000649110
29. ^Swain, Gyana. "Meta hits pause on 'Llama 4 Behemoth' AI model amid capability concerns." Computerworld, May 16, 2025. computerworld.com/...odel-amid-capability-concerns
30. ^"Meta chief AI scientist Yann LeCun plans to exit to launch startup, FT reports." Reuters (Yahoo Finance syndication), November 11, 2025. finance.yahoo.com/...f-ai-scientist-yann-114034245
31. ^1 ^2 ^3 ^4 ^5 ^6Meta. "Introducing Muse: The World's First Personal AI Agent Built for Everyone." Meta Newsroom, September 8, 2026. about.fb.com/...introducing-muse-personal-ai-agent
32. ^Sheasha, Tarek. "How We Built Safety Into Muse." Meta AI Research, September 8, 2026. research.meta.ai/...-agents-our-approach-with-muse
33. ^1 ^2 ^3"Introducing Muse Spark 1.3." Meta Superintelligence Labs, September 2, 2026. research.meta.ai/...introducing-muse-spark-1-3
34. ^1 ^2Alexandr Wang (@alexandr_wang). "1/ we just publicly released Muse Spark 1.3 max!" X post, September 4, 2026. x.com/...2095938990197329935
35. ^1 ^2Muse (@Muse). "Made it to Canada. Where to next?" X post, September 18, 2026. x.com/...2101071767838372272
36. ^Muse (@Muse). "Available on iOS and Web. Coming soon for Android." X post, September 18, 2026. x.com/...2101071769448935749
37. ^Mark Zuckerberg (@finkd). "Muse for Mac is out today!" X post, September 17, 2026. x.com/...2100713341555712149
38. ^1 ^2 ^3 ^4Vanian, Jonathan. "Meta's Muse AI agent downloads are surging. Here's how it compares to ChatGPT, Grok and Claude." CNBC, September 21, 2026. cnbc.com/...meta-muse-personal-ai-agent-downloads
39. ^Perez, Sarah. "Meta's Muse is outpacing ChatGPT's early mobile launch." TechCrunch, September 21, 2026. techcrunch.com/...ing-chatgpts-early-mobile-launch
40. ^Muse (@Muse). "icymi: 4 new connectors announced in the last 24 hours, all coming soon." X post, September 23, 2026 (UTC). x.com/...2102603447791170034
41. ^1 ^2 ^3Weatherbed, Jess. "Meta patches Muse exploit that let attackers control the AI agent." The Verge, September 22, 2026. theverge.com/...se-patch-zero-day-exploit-ai-agent
42. ^1 ^2 ^3Paul, Katie. "Meta testing a 'human concierge' for its new personal AI agent, Muse." Reuters (BNN Bloomberg syndication), September 22, 2026. bnnbloomberg.ca/...ai-agent-muse-reuters-exclusive
43. ^1 ^2 ^3"Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device." Meta Superintelligence Labs, August 10, 2026. research.meta.ai/...use-glimmer-open-agentic-model
44. ^Wheatley, Mike. "Meta Platforms creates new organization to lead its AI infrastructure buildout." SiliconANGLE, January 12, 2026. siliconangle.com/...ead-ai-infrastructure-buildout
45. ^"OpenAI hires ex-Apple models head from Meta, The Information reports." Reuters (Yahoo Finance syndication), February 25, 2026. finance.yahoo.com/...-meta-ai-researcher-020959357
46. ^Albergotti, Reed. "AI researcher Andrew Tulloch is leaving Meta." Semafor, September 9, 2026. semafor.com/...cher-andrew-tulloch-is-leaving-meta
47. ^Bastian, Matthias. "Meta acqui-hires Dreamer's entire team to bolster its lagging AI agent ambitions." The Decoder, March 23, 2026. the-decoder.com/...-its-lagging-ai-agent-ambitions
48. ^Fried, Ina. "Meta hires OpenAI veteran Luke Metz." Axios (Yahoo Finance syndication), August 24, 2026. finance.yahoo.com/...openai-veteran-luke-031156286
49. ^1 ^2AI at Meta (@AIatMeta). "Introducing Muse, a personal agent that gets things done for you, powered by Muse Spark 1.3." X post, September 8, 2026. x.com/...2097401493770956808
50. ^"Introducing Muse Spark 1.3." Meta Superintelligence Labs, as archived by the Internet Archive on September 2, 2026. web.archive.org/...introducing-muse-spark-1-3
51. ^Song, Dawn (@dawnsongtweets). Post on X announcing her move to Meta Superintelligence Labs. June 25, 2026. x.com/...2070191051873345910
52. ^Hale, Craig. "'The goal is not to replace humans': new Meta AI research chief Dawn Song says the next frontier is AI agents that are 'economically valuable'." TechRadar, June 29, 2026. techradar.com/...ts-that-are-economically-valuable