Category: Business

Meta launched Muse Spark on Wednesday, marking the first model built by Meta Superintelligence Labs—the team assembled nine months ago under Chief AI Officer Alexandr Wang after Meta’s $14 billion Scale AI acquisition. It’s live now at meta.ai and the Meta AI app, with a rollout to Facebook, Instagram, and WhatsApp coming in the next few weeks.

This isn’t just another chatbot upgrade or a new version of Llama. Muse Spark is natively multimodal—it processes images, text, and voice from the ground up, rather than bolting vision onto an existing text model. It comes with visual chain-of-thought, tool-use support, and something Meta is calling “Contemplating mode”: a setup that runs multiple AI agents in parallel to tackle harder problems. That’s Meta’s answer to the extended thinking modes from Google’s Gemini Deep Think and OpenAI’s GPT Pro.

“Muse Spark is the first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts,” Meta wrote in an official announcement. “To support further scaling, we are making strategic investments across the entire stack—from research and model training to infrastructure, including the Hyperion data center.”

The company worked with more than 1,000 physicians to curate training data for Muse Spark’s medical reasoning. The results on HealthBench Hard—an open-ended health queries benchmark—are striking: Muse Spark scored 42.8, compared to 40.1 for GPT 5.4 and just 20.6 for Gemini 3.1 Pro. That’s not a marginal difference.

On agentic search (DeepSearchQA), Muse Spark also leads with 74.8, beating Gemini (69.7) and GPT 5.4 (73.6). On CharXiv Reasoning—figure understanding from scientific papers—it scored 86.4, the highest across the models in the comparison.

For those into jailbreaking AI, the model was cracked open within minutes:

But good isn’t the same as great. The overall benchmark picture shows Gemini 3.1 Pro still running ahead on most categories. The gap is most visible on ARC AGI 2, the abstract reasoning puzzle benchmark: Gemini scored 76.5 to Muse Spark’s 42.5.

On coding (LiveCodeBench Pro), Gemini’s 82.9 outpaces Meta’s 80.0. On MMMU Pro—multimodal understanding—Gemini scored 83.9 versus 80.4. Meta’s own blog acknowledges current performance gaps in long-horizon agentic systems and coding workflows.

There’s also a notable strategic shift baked into this launch. Muse Spark is a closed model—its architecture and weights won’t be made public. That’s a sharp departure from Llama, which built Meta’s reputation in open AI circles. After Llama 4’s underwhelming reception earlier this year, Meta appears to have decided the next chapter needs to be written differently.

The company says it hopes to open-source future versions of Muse, but for now the code stays inside Meta. The tech giant’s stock climbed nearly 9% on Wednesday following the announcement, and finished the trading day up 6.5% to a price of $612.42.

“Contemplating mode” uses parallel agent orchestration to push the model’s ceiling higher. In that configuration, Muse Spark hit 58% on Humanity’s Last Exam and 38% on FrontierScience Research—territory that makes it competitive with the most capable versions of Gemini and GPT, rather than their standard releases.

Meta is also rolling out a shopping assistant that compares products and links directly to purchases, and plans to bring Muse Spark to Facebook, Instagram, and WhatsApp in the coming weeks—following the same script implemented since Llama 3, putting it in front of more than 3.5 billion users. A private API preview is opening to select developers.

The model was built in nine months, internally codenamed Avocado, with Meta claiming that its new pretraining stack can reach the same capability level as Llama 4 Maverick using over 10 times less compute.

Muse Spark is described internally as a “small and fast” first step in the Muse family. A more capable version is already in development.

Publicly traded Bitcoin miner Cango Inc. cut its average production cost by 19.3% to $68,216 per BTC in March—down from $84,552 in Q4 last year—achieving the reduction through strategic fleet optimization rather than expansion.

The company decommissioned older mining hardware and relocated operations to regions with cheaper power, while selling 2,000 Bitcoin during the month to retire crypto-backed debt. That tally of Bitcoin is currently valued around $143 million, and the firm used the proceeds to trim its outstanding loan balances to $30.6 million.

Cango still held 1,025.69 BTC in its treasury as of the end of March 31, valued over $73 million as of this writing. The firm’s total hash rate stood at 37.01 EH/s as of the end of March, split between 27.98 EH/s from self-mining and 9.02 EH/s from leasing arrangements.

The operational restructuring involved more than simple downsizing. In high-cost hosting locations, Cango deployed hash rate leasing models to maintain revenue without bearing full operational expenses, according to the company’s announcement.

Cango plans to redirect capital from its deleveraging efforts toward AI computing infrastructure, positioning the cost reductions as preparation for business model expansion. The same filing indicated the company views AI infrastructure as a natural extension of its existing power and facility investments.

The efficiency focus reflects shifting priorities among public Bitcoin miners facing compressed margins and market volatility. Rather than competing solely on hash rate growth, companies are examining unit economics and alternative revenue streams. Several Bitcoin mining firms have made moves into powering AI computing needs, even abandoning their original business focuses in an effort to chase larger profits amid the AI boom.

Cango’s operational restructuring follows similar moves across the public mining sector. MARA recently sold $1.1 billion in Bitcoin to buy back convertible debt while cutting 15% of its workforce. Core Scientific has explored plans to sell all of its Bitcoin holdings to finance its own AI transition, while Cipher Digital shifted focus to data center operations with a 15-year infrastructure deal, highlighting the industry’s evolution beyond traditional mining models.

Cango shares (CANG) finished the trading day up 3.3% on Wednesday at a price of $0.4291 on a broadly green day for stocks, following a conditional ceasefire between the U.S. and Iran. Despite the daily uptick, however, CANG shares have fallen nearly 39% in the last month.

Aiming to address the rise of AI-enabled child sexual exploitation, OpenAI on Wednesday published a policy blueprint outlining new safety measures the industry can take to help curb the use of AI in creating child sexual abuse material.

In the framework, OpenAI lists legal, operational, and technical measures aimed at strengthening protections against AI-enabled abuse and improving coordination between technology companies and investigators.

“Child sexual exploitation is one of the most urgent challenges of the digital age,” the company wrote. “AI is rapidly changing both how these harms emerge across the industry and how they can be addressed at scale.”

OpenAI said the proposal incorporates feedback from organizations working in child protection and online safety, including the National Center for Missing and Exploited Children and the Attorney General Alliance and its AI task force.

“Generative AI is accelerating the crime of online child sexual exploitation in deeply troubling ways-lowering barriers, increasing scale, and enabling new forms of harm,” President & CEO, National Center for Missing & Exploited Children, Michelle DeLaune said in a statement. “But at the same time, the National Center for Missing & Exploited Children is encouraged to see companies like OpenAI reflect on how these tools can be designed more responsibly, with safeguards built in from the start.”

OpenAI said the framework combines legal standards, industry reporting systems, and technical safeguards within AI models. The company said these measures aim to help identify exploitation risks earlier and improve accountability across online platforms.

The blueprint identifies areas for action, including updating laws to address AI-generated or altered child sexual abuse material, improving how online providers report abuse signals and coordinate with investigators, and building safeguards into AI systems designed to prevent misuse.

“No single intervention can address this challenge alone,” the company wrote. “This framework brings together legal, operational, and technical approaches to better identify risks, accelerate responses, and support accountability, while ensuring that enforcement authorities remain strong as technology evolves.”

The blueprint comes as child safety advocates have raised concerns that generative AI systems capable of producing realistic images could be used to create manipulated or synthetic depictions of minors. In February, UNICEF called on world governments to pass laws criminalizing AI-generated child abuse material.

In January, the European Commission launched a formal investigation into whether X, formerly known as Twitter, violated EU digital rules by failing to prevent the platform’s native AI model, Grok, from generating illegal content, as regulators in the United Kingdom and Australia have also opened investigations.

Noting that laws alone will not stop the scourge of AI-generated abuse material, OpenAI said stronger industry standards will be necessary as AI systems become more capable.

“By interrupting exploitation attempts sooner, improving the quality of signals sent to law enforcement, and strengthening accountability across the ecosystem, this framework aims to prevent harm before it happens and help ensure faster protection for children when risks emerge,” OpenAI said.

The race between quantum computers and Bitcoin’s cryptography has become a recurring theme in the cryptocurrency industry. But even as anxieties over Bitcoin’s “Q-day” grow, a new report from investment firm Bernstein says the outcome is unlikely to be catastrophic for the world’s largest cryptocurrency.

Instead, the firm argues quantum computing should be treated as a long-term upgrade cycle for Bitcoin and the broader crypto industry rather than an existential threat to the network.

“The risk is neither existential, nor novel, and also not limited to crypto,” Bernstein wrote, noting that quantum computing also posed a threat to everything from financial services, military, and healthcare.

According to Bernstein, the highest threat from quantum computing is to the 1.7 million BTC, around $116.6 billion, in legacy wallets from the days when Satoshi Nakamoto was still active online. That’s because this stash of Bitcoin was stored in early address formats that expose public keys on the blockchain and could be targeted in a “harvest now, decrypt later” attack. For newer encryption protocols, chains, and crypto-linked real-world assets, the threat is limited to some unsafe practices that can be mitigated and managed, the firm said.

Bernstein also emphasized that quantum computing won’t impact Bitcoin mining in the near future.

“Bitcoin mining has no realistic risk from [quantum computers] based on Shor’s algorithm, as SHA encryption used in mining is quantum safe—several millions of years even after recent improvements, including Grover’s algorithm.”

Blockstream CEO Adam Back, a Bitcoin pioneer, who was recently named as the likely person behind the identity of Satoshi Nakamoto according to a new report by The New York Times, expressed a similar view.

“The Google paper is talking about algorithmic improvements, and doesn’t bring with it any hardware improvements,” Back told Bloomberg on Tuesday.

Back’s comments come as concern over quantum computing intensified after new academic research suggested fewer quantum resources may be needed to break elliptic-curve cryptography, the digital signature system used by Bitcoin wallets. A March paper from Google Quantum AI also shortened estimates for when such capabilities could emerge, pointing to a possible timeline around 2032.

Current quantum computers operate with roughly a thousand physical qubits. Breaking the cryptography used by Bitcoin would require hundreds of thousands of stable, error-corrected qubits along with major advances in engineering and hardware reliability.

Back said current quantum systems remain “extremely basic” because of limitations with error correction, calling even the most advanced demonstrations trivial compared with the calculations needed to compromise Bitcoin’s cryptography.

“The biggest calculation it’s performed is that to factorize the number 21 into seven times three,” he said. “Sort of thing that primary school children can do.”

Bitcoin relies on elliptic-curve cryptography to secure transactions and SHA-256 hashing to power mining. While the Bernstein report suggests that quantum computers could eventually target the signature system, they are unlikely to threaten the mining algorithm.

The best approach, Back said, is to prepare Bitcoin users for a gradual transition to quantum-resistant security.

“The prudent thing to do is to prepare Bitcoin and give people the option to migrate their keys to a quantum-ready format,” he said. “The longer time that Bitcoin users have in order to migrate their keys for custodians and exchanges to move their coins to a quantum-ready format, the safer it will be,” he said.

A company partnered with World Liberty Financial—the crypto project backed by President Donald Trump—has been linked to individuals connected to a sanctioned Cambodian conglomerate accused of running global scam operations, according to an investigation published Monday by the Organized Crime and Corruption Reporting Project and Guardian Australia.

The investigation found that a planned “blockchain theme resort” in Timor-Leste tied to the partner, AB network, involved three individuals who were later sanctioned by the U.S. Treasury as part of a crackdown on the Cambodia-based Prince Group. The project, promoted as a luxury destination for cryptocurrency innovators, was backed by a local development company established with $10 million in capital.

Corporate records show that the majority shareholder of the development company was Yang Jian, a Cyprus-based businessman sanctioned in October for allegedly working with Prince Group CEO Chen Zhi on a separate resort project described by U.S. authorities as a “predatory investment.”

The three sanctioned individuals were removed from the Timor-Leste project shortly after the sanctions were announced, and there is no evidence that illicit funds flowed into the development, or that AB network is directly connected to the Prince Group.

The AB network announced its partnership with World Liberty Financial in November, granting it the right to use the company’s U.S. dollar-pegged stablecoin USD1 on its blockchain. The collaboration followed a series of high-profile announcements by AB, which has promoted ties to global political figures and listed former world leaders among advisers to its Irish-registered nonprofit arm.

World Liberty Financial, founded in 2024 by partners including companies affiliated with the Trump and Witkoff families, said it had carried out due diligence on AB and was not made aware of the resort, or of individuals linked to the Timor-Leste project. Lawyers for the company told investigators it is “committed to responsible practices and compliance” and said claims of links to sanctioned figures are “unfounded and untrue.”

According to the FBI’s 2025 Internet Crime Report, Americans alone lost nearly $21 billion to online scams last year, with more than 1 million complaints filed. Cryptocurrency-related fraud accounted for the largest share of losses, totaling more than $11 billion across 181,565 complaints.

The Prince Group, a Cambodia-based conglomerate led by Chen Zhi, has been accused by U.S. authorities of operating one of the world’s largest online scam networks, allegedly generating tens of billions of dollars annually through fraud schemes run out of compounds across Southeast Asia.

The U.S. government last year seized $15 billion worth of Bitcoin from Chen in what it described as its largest forfeiture action against online scammers. The company has denied wrongdoing. Cambodian authorities arrested and then extradited Chen to China in January.

The AB network’s corporate structure has remained opaque. It describes itself as a decentralized ecosystem comprising an Irish nonprofit, a Cayman Islands foundation, and blockchain-based entities governed by token holders. The OCCRP‘s reporting identified two previously undisclosed figures—software developer Sui Chenggang and entrepreneur Lin Xiaofan—as key actors within the network.

Lin, who said he played a leading role in the Timor-Leste resort project, denied any connection to the Prince Group. He also said he introduced Sui to World Liberty executives.

Promotional material for the Timor-Leste development was removed from AB’s websites after OCCRP reporters began making inquiries. Lin said the project remains active but that AB is no longer involved, and provided documentation showing the partnership was terminated in November.

Despite distancing itself from the sanctioned individuals, AB continues to promote its partnership with World Liberty Financial and its network of political advisers. The collaboration has so far produced limited uptake of World Liberty’s stablecoin on AB’s blockchain, with a max total supply of around $3.6 million and just over 3,000 holders.

World Liberty Financial did not respond to additional requests for comment.

Editor’s note: This story was updated after publication to move up context about the partnership.