Tag: Business – Decrypt

The CIA recently used AI to generate an intelligence report without a human analyst driving it. Deputy Director Michael Ellis confirmed the milestone Thursday at a Special Competitive Studies Project event, marking a shift from quiet experimentation to a public declaration of ambition.

Ellis said the agency ran more than 300 AI projects last year, Politico reports. Somewhere in that stack, a machine produced an intelligence product entirely on its own—a first in the agency’s history.

The near-term roadmap is more incremental. Analysts would get AI “coworkers” embedded in agency analytics platforms to handle drafting, editing for clarity, and benchmarking outputs against tradecraft standards. Humans would still ultimately sign-off on the results. But the goal is speed—getting intelligence products out faster than a human-only pipeline allows.

Within a decade, Ellis said, CIA officers will manage teams of AI agents operating as “autonomous mission partners,” a hybrid model that scales intelligence gathering in ways no human workforce can match alone.

The CIA has been building toward this for years. In 2023, the intelligence agency announced its own AI chatbot to help staffers parse surveillance data. By 2024, CIA Director Bill Burns and MI6 Chief Richard Moore jointly disclosed they were actively using generative AI for content triage, analyst support, and tracking how foreign adversaries deploy the technology. Ellis’ remarks push that public timeline forward considerably.

Earlier this year, Anthropic declined to relax restrictions barring its tools from domestic surveillance or fully autonomous weapons applications. Defense Secretary Pete Hegseth responded by designating Anthropic’s products a “supply chain risk.” President Trump then ordered every federal agency to phase out Anthropic tools. The company has legally challenged the move.

Ellis didn’t name Anthropic, but the message landed clearly. The CIA “cannot allow the whims of a single company” to constrain its use of AI, he said, and the agency is actively diversifying across vendors to stay operationally flexible.

Ellis also flagged that the CIA doubled its technology-focused foreign intelligence reporting, tracking how adversaries like China are deploying AI across semiconductors, cloud computing, and R&D. The agency’s Center for Cyber Intelligence was elevated to a full mission center—a move Ellis described as critical, given that “the battle of cybersecurity will be a battle of artificial intelligence.”

U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell reportedly convened a meeting with Wall Street bank CEOs earlier this week to warn about cybersecurity risks tied to a new artificial intelligence model from Anthropic.

According to a report by Bloomberg, the meeting included executives from Citigroup, Bank of America, Wells Fargo, Morgan Stanley, and Goldman Sachs. Officials discussed Anthropic’s new AI model Mythos, which has recently drawn broad concern over its apparent advanced cybersecurity capabilities.

Officials convened the meeting to ensure banks understand the risks posed by systems capable of identifying and exploiting software vulnerabilities across operating systems and web browsers, and to encourage institutions to strengthen defenses against potential AI-assisted cyberattacks targeting financial infrastructure.

Security researchers have warned that tools capable of automatically discovering vulnerabilities could accelerate both defensive security work and malicious hacking if misused.

Anthropic’s Mythos model first surfaced online in March after draft materials about the system leaked online, revealing what the company described as its most capable AI model yet. In testing, the system reportedly found thousands of previously unknown software vulnerabilities, including zero-day flaws across major operating systems and web browsers.

Anthropic researchers said in a report earlier this week that Mythos Preview’s vulnerability-discovery capabilities were not intentionally trained, but instead emerged from broader improvements in the model’s coding, reasoning, and autonomy.

“The same improvements that make the model substantially more effective at patching vulnerabilities also make it substantially more effective at exploiting them,” the firm wrote.

Because of those capabilities, Anthropic has restricted access to a small group of cybersecurity organizations.

“Given the strength of its capabilities, we’re being deliberate about how we release it,” Anthropic said in a statement. “As is standard practice across the industry, we’re working with a small group of early access customers to test the model. We consider this model a step change and the most capable we’ve built to date.”

To address that risk, Anthropic is testing Mythos through Project Glasswing, a collaboration with major technology and cybersecurity companies that uses the model to identify and patch vulnerabilities in critical software before attackers can exploit them.

“Project Glasswing is a starting point. No one organization can solve these cybersecurity problems alone,” the company said in a statement. “Frontier AI developers, other software companies, security researchers, open-source maintainers, and governments across the world all have essential roles to play.”

Anthropic did not immediately respond to Decrypt’s request for comment.

Exodus, the publicly traded crypto wallet provider, began rolling out a new “Exodus Pay” feature on Wednesday, aiming to turn its self-custodial storage app into a tool for everyday payments. The launch is currently limited to users in five states, including New York and California.

The Omaha-based firm listed its stock on the New York Stock Exchange in 2024 and says the new feature expands the role of its wallet beyond storage into payments. The company positions the service as an alternative to centralized payment apps.

“Most payment apps are third parties that hold your funds for you,” Exodus co-founder and CEO JP Richardson told Decrypt. “That means they can freeze your account, reverse transactions, and decide what you’re allowed to buy.” Exodus, by contrast, can’t do that, because users remain in control of their funds at all times.

The company says Exodus Pay works within the existing wallet app and allows users to spend USD-backed stablecoins, such as USDC, or Bitcoin at merchants that accept Visa or Apple Pay.

“The problem with self-custody until now has been the friction. Seed phrases, complicated networks—most self-custody consumer experiences aren’t built for someone who just wants to pay for groceries or send friends money,” Richardson said.

To encourage adoption, Exodus says it will subsidize network fees and allow transfers using phone numbers. The service remains geographically limited due to regulatory requirements. It is currently available only in Nebraska, Texas, Florida, New York, and California.

Richardson said the company has focused on simplifying the user experience, claiming that “someone with zero crypto experience should be able to use an app intuitively.”

Exodus joins a growing list of crypto wallet developers that let customers pay for purchases using crypto or stablecoins, including Coinbase, BitPay, and PayPal.

After the initial launch, Richardson said the company plans to expand the service nationwide over the next several weeks.

“By mid-April, everyone in America will have Exodus Pay in their app,” Richardson said. For existing users, the feature will appear as an automatic update rather than a new download. “If you already have Exodus, you’ll have Exodus Pay,” he added.

Crypto firms like Coinbase and Binance, alongside government agencies like the United States Secret Service and the U.K.’s National Crime Agency (NCA), have flagged $45 million in stolen crypto funds as part of fraud schemes, the parties announced on Thursday. 

In the probe, more than 20,000 victims of approval phishing fraud were identified, and $12 million in funds were frozen in the hopes of returning funds to victims.

“To take on approval phishing at scale, our Global Intelligence team joined forces with multiple international law enforcement agencies and other partners for a focused operational sprint held at the National Crime Agency’s headquarters in London,” Coinbase wrote. 

“The goal was straightforward: identify victims, trace stolen funds, and disrupt the infrastructure that makes approval phishing possible—as fast as we could,” it added. 

The investigative sprint, dubbed “Operation Atlantic,” was first revealed last month and was hosted by the NCA at its headquarters in London. In a week of focused work there, the agencies disrupted “multiple fraud networks,” and will continue to analyze intelligence gathered moving forward. Other crypto firms, like on-chain security firm Chainalysis, crypto exchange Kraken, and stablecoin issuer Tether, were included as partners. 

“Operation Atlantic is a powerful example of what is possible when international agencies and private industry work side by side,” said National Crime Agency Deputy Director of Investigations Miles Bronfield, in a statement. “This intensive action has led to the safeguarding of thousands of victims in the UK and overseas, stopped criminals in their tracks and helped save others from losing their funds.”

The enforcement campaign was focused on crypto investors who may have been impacted by approval phishing, when malicious actors attempt to gain access to funds via fake pop-up notifications or alerts that unsuspecting victims believe come from trusted parties. 

More than 120 web domains used for schemes were identified during the week, according to the Secret Service. 

“With traditional financial crimes, this kind of cross-border, multi-agency coordination would take months,” Coinbase wrote in its recap on the week. “With blockchain technology, we moved from identification to action in a single week-long sprint.” 

The engagement report comes just over a week after alleged North Korean hackers made off with around $285 million via an exploit of Solana protocol, Drift. The exploit would represent just a fraction of the funds lost to crypto scam last year, with a recent report from the FBI indicating that more than $11.4 billion was lost to crypto scams in 2025 alone.

Morning Minute is a daily newsletter written by Tyler Warner. The analysis and opinions expressed are his own and do not necessarily reflect those of Decrypt. And check out our new daily news show covering all of the top stories in 5 minutes or less, downloadable on Apple Pod or Spotify.

GM!

Today’s top news:

• Crypto majors up 1%; BTC at $72K
• ZEC and MON rally 20% leading top movers; HYPE +5%
• CZ and OKX founder Star publicly dispute, CZ calls for $1B “divorce” bet
• Bessent and Brian Armstrong both say it’s time to pass the Clarity Act
• WLFI falls 10% after team reveals using 5B tokens to borrow $75M

🌎 Bitcoin breaks $73K on ceasefire hopes, STRC

Bitcoin topped $73,000 briefly on Thursday, per data from CoinMarketCap, reversing an early sell-off after Netanyahu signaled Lebanon negotiations. It bounced off that level and is holding just above $72K this morning.

The Bitcoin options market is even more bullish. Data shows that the $80,000 level is seeing the most volume in June expiry contracts with over $1.6B in open interest, a full 10% move from current levels.

As for why traders are bullish—well, it could be Saylor-driven. Strategy‘s STRC had another massive day Thursday with over 3M preferred shares moved, generating capital to purchase 2,000+ Bitcoin ($144M). Wednesday’s numbers were similar, and the totals historically rise into the dividend cutoff date (next Wednesday). So expect 3 more days of increasing STRC flows.

It’s a strong setup for Bitcoin near-term.

Key Details:

• Bitcoin topped $73,000 Thursday, up ~9% over the past month as crypto decouples from software stocks, which are down 12% over the same stretch
• June $80K Bitcoin options showing $1.6B in open interest, most concentrated target
• Saylor’s STRC moved 3M shares, enough to buy $144M in Bitcoin; on pace for well over $300M in purchases this week

📊 Galaxy profit rockets, stock jumps

The headline number from Galaxy’s 2025 annual report, a $241M net loss, buried the more important one: The firm’s Digital Assets segment generated $505M in adjusted gross profit.

GLXY closed up 11.3% Thursday, second-best crypto equity on the day.

The thesis Mike Novogratz is selling isn’t a crypto trading story anymore. It’s AI infrastructure. Galaxy’s Helios campus, once one of North America’s largest Bitcoin mines, is an 800-megawatt facility fully leased to CoreWeave that’s beginning to generate compute revenue in 2026. “The most consequential shift right now is the move from narrative to infrastructure,” he wrote.

That pivot from BTC mining to AI is clearly paying off…

Key Details:

• Galaxy posted a $241M 2025 net loss driven by unrealized losses and one-time costs; core Digital Assets segment generated $505M in adjusted gross profit
• GLXY closed up 11.3% at $21.15; total assets on platform hit $12B with $2B in net inflows during 2025
• The Helios play: 800MW Texas facility fully leased to CoreWeave; AI compute revenue begins 2026; Galaxy is pitching itself as half crypto financial firm, half AI infrastructure company

🦅 Gemini is on sale, but nobody wants the whole thing

Potential buyers are circling Gemini, but not in the way the Winklevoss twins might want.

Per CoinDesk, interested parties are evaluating an acquisition of Gemini’s shuttered EU and UK operations specifically to obtain MiCA and FCA regulatory licenses. Nobody is pursuing a full takeover.

The backdrop is stark. Gemini IPO’d at $28 in September 2025 and now trades around $4.70, down 83%. The company cut 25% of its workforce in February, exited the EU, UK, and Australia, lost three senior executives, and faces a shareholder class-action lawsuit filed in March.

GEMI stock jumped 11% on the acquisition reports, but has already shed some of those gains.

Key Details:

• Potential buyers are circling Gemini’s shuttered EU and UK operations for MiCA and FCA licenses; no full takeover interest
• The distress context: $28 IPO September 2025, now $4.70 (down 83%); 25% workforce cut; exited EU/UK/Australia; three executives departed; class-action lawsuit filed March 2026
• MiCA wrinkle: license doesn’t transfer in an acquisition; change of control triggers full regulatory reassessment; buyers face scrutiny equivalent to a new applicant

⚖️ Bessent to the Senate: Pass the Clarity Act

Treasury Secretary Bessent made his most direct push yet Thursday, urging the Senate to pass the Clarity Act and resolve the stablecoin yield dispute still stalling the bill.

This comes just one day after the White House Council of Economic Advisors mathematically dismantled the banking lobby’s core argument, finding a yield ban would boost lending by just $2.1B, a 0.02% increase.

The only remaining variable is whether Senate Democrats and holdout Republicans will accept a stablecoin yield framework that Coinbase can live with.

And we may have gotten a signal from Brian Armstrong last night, who tweeted “It’s time to pass the Clarity Act” in union with Bessent.

Key Details:

• Bessent urged the Senate to pass the Clarity Act, calling for resolution on stablecoin yield provisions; follows the White House CEA report that undercut the banking lobby’s deposit-flight argument
• What’s left: Senate Banking Committee markup; stablecoin yield language is the last unlock for the full US crypto regulatory stack
• Odds of the Clarity Act passing in 2026 rose 3% to 59% on Thursday

🤖 Florida goes after OpenAI

Florida AG James Uthmeier launched a formal investigation into OpenAI and ChatGPT Thursday, citing the chatbot’s alleged role in the April 2025 FSU mass shooting that killed two people, child safety concerns, and the risk of OpenAI data reaching the Chinese government.

The quote Uthmeier posted to announce it is the week’s most ironic AI headline: “AI should advance mankind, not destroy it.”

The investigation arrives as AI infrastructure, specifically data centers, are coming under attack. Per Bloomberg and Sightline Climate, 30-50% of the data centers planned to come online this year are facing delays or outright cancellations. Of the 12 gigawatts of capacity announced for 2026, only a third is currently under construction. Bernie Sanders and AOC introduced the AI Data Center Moratorium Act in March to stop all new construction until federal safeguards are in place. It’s not going anywhere, but it signals the political mood around AI is shifting.

Key Details:

• Florida AG Uthmeier launched a formal OpenAI investigation, citing ChatGPT’s alleged role in the 2025 FSU shooting, child safety, and CCP data concerns; subpoenas forthcoming; arrives as OpenAI eyes a $1T IPO
• 30-50% of US data centers planned for 2026 are facing delays or cancellations per Sightline Climate
• The political pressure: Sanders and AOC introduced the AI Data Center Moratorium Act in March, calling for a full construction halt until federal safeguards are in place; fringe bill, real signal

🌎 Macro crypto and markets

• Crypto majors are slightly green; BTC +1% at $72.1k; ETH +1% at $2,210; SOL +2% at $84; HYPE +5% at $41
• DEXE (+30%), ZEC (+20%), and MON (+20%) led top movers
• Oil -3% at $94; Gold even at $4,764
• The Bitcoin options market is showing concentration at the $80,000 price level for June expiry contracts with over $1.6B in open interest
• Treasury Secretary Bessent urged the Senate to pass the Clarity Act, pushing for resolution on stablecoin yield provisions still stalling the bill
• The Treasury will share cybersecurity intelligence with crypto firms, giving the industry access to the same threat data distributed to traditional financial institutions
• Former SEC official Brett Redfearn joined Securitize as president ahead of the BlackRock-backed tokenization firm’s anticipated public listing
• Binance founder CZ got into a public dispute with OKX founder Star, escalating to the point that CZ bet Star $1B that he is “officially divorced” from Binance

Corporate Treasuries & ETFs

Meme Coin Tracker

• Meme leaders were slightly green; DOGE +1%, SHIB +1%, PEPE +1%, TRUMP -2%, PENGU +4%, SPX +5%, FARTCOIN +2%
• SBTI (25x), triplet (+102%), chillguy (+32%), and hodl (+38%) led notable onchain movers

💰 Token, airdrop & protocol Tracker

• Tether released its QVAC SDK, a toolkit enabling AI apps to run locally on devices without cloud servers, extending Tether’s push into AI infrastructure
• Nunchuk released open-source tools letting AI agents interact with Bitcoin wallets via multi-sig, without giving agents unilateral control over funds
• DeFi lender Sky is restructuring its products to pursue a formal credit rating, targeting institutional capital as DeFi protocols push further into TradFi
• Binance enabled prediction market trading in-app via Predict.fun, giving 240M+ users direct access to event contracts as the CFTC battles states over federal jurisdiction
• WLFI fell 10% after the team revealed that it borrowed $75M against 5B tokens

🚚 What is happening in NFTs?

• NFT leaders were mostly flat again; Punks -1% at 28 ETH, Pudgy -1 at 4.2 ETH, BAYC even at 6.39 ETH; Hypurr’s +!% at 392 HYPE
• Nouns (+69%) and Kodas (+13%) led notable movers

OpenAI is currently building a cybersecurity product it plans to release exclusively through its “Trusted Access for Cyber” program, according to Axios. The program was previously announced in February, and it’s meant to be a controlled rollout that keeps certain products away from the general public and in the hands of defensive security operators only.

OpenAI launched the program after releasing GPT-5.3-Codex, currently its most capable cybersecurity offering, and is backing participant access with $10 million in API credits.

The news comes amid growing worry among cybersecurity experts over the potential for increasingly powerful AI products overwhelming existing systems. Just earlier this week, Anthropic spooked itself with its own creation, Claude Mythos.

Anthropic said Mythos is the company’s most capable AI model, and turned out to be so effective at finding security vulnerabilities—zero-days in every major operating system and browser—that it decided only a handpicked group of organizations should have access to it.

Now OpenAI is, reportedly, doing something similar.

Anthropic is currently fighting a legal battle after the Pentagon designated it a supply chain risk after the company refused to lift usage restrictions on Claude for surveillance and autonomous weapons applications. Federal agencies have been scrutinizing AI companies’ safety protocols with increasing intensity since early April.

As of now, OpenAI has not shared any public information officially confirming or denying the reports.

The reason for the restrictions isn’t subtle. Anthropic’s Mythos Preview, which leaked before its official rollout, was found capable of identifying “tens of thousands of vulnerabilities” that even advanced human bug hunters would struggle to locate. The model is described as “extremely autonomous” and reasons with the sophistication of a senior security researcher. That kind of capability, available to anyone with an API key, is the kind of thing that keeps security teams up at night.

Anthropic’s response was Project Glasswing—a controlled access initiative that gives Mythos Preview only to vetted organizations: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, Palo Alto Networks, and roughly 40 others involved in maintaining critical infrastructure.

OpenAI’s decision to lock down products like this one looks like an attempt to get ahead of that regulatory pressure. By voluntarily restricting access before a government agency tells them to, OpenAI positions itself as the responsible actor in a space where Anthropic is getting hammered.

The restrictions also reflect something deeper than caution about one specific model. Anthropic’s own safety report acknowledged that Cybench, the benchmark used to evaluate whether an AI poses serious cyber risk, “is no longer sufficiently informative of current frontier model capabilities”—because Mythos cleared it completely. The tool built to measure the danger is no longer adequate for what’s being built. Anthropic added that its overall safety determination “involves judgment calls” and that many evaluations leave “more fundamental uncertainty.”

Anthropic committed up to $100 million in usage credits and $4 million in direct donations to open-source security organizations as part of its rollout. OpenAI has not announced a comparable commitment alongside its access program, though both companies are framing their restricted programs as a net benefit for defensive security—the idea being that giving better tools to defenders before attackers get them is worth the tradeoff of limiting general access.

The pattern emerging across the frontier AI industry is that the most capable models will no longer arrive as broad product launches. They’ll be distributed more like classified research—selectively, under agreement, to organizations with the infrastructure and intent to use them responsibly.

Canary Capital thrust Pepe into the limelight on Wednesday with an application for an exchange-traded fund that tracks the meme coin’s price, but the token’s muted reaction may serve as the latest sign of Wall Street’s tepid appetite for assets that trade on vibes.

On Thursday, Pepe changed hands around $0.00000359, up about 0.6% over the last day, according to CoinGecko. The day before, trading volume rose 10% to $432 million.

Not long ago, meme coins served as key growth drivers for firms like Wintermute. Yet the crypto market maker acknowledged last year that its prediction of a core asset manager debuting a meme coin ETF, particularly Dogecoin, was intended to be tongue-in-cheek.

Today, four crypto asset managers offer U.S.-listed Dogecoin ETFs. Still, it remains “very hard for institutional investors to construct a credible investment rationale around something like Doge, which is perhaps more geared towards the retail audience,” James Butterfill, head of research at crypto asset manager CoinShares, told Decrypt.

Dogecoin is ranked 17th out of all crypto ETFs that CoinShares tracks, generating $13 million worth of year-to-date inflows. Outside of ETFs tracking Bitcoin, Ethereum, Solana, and XRP, Butterfill noted that ETFs tied to other altcoins represent 9% of total assets under management.

“They’re just not popular with investors,” he said. “It’s the big four and not much else.”

Decrypt has reached out to Canary for comment.

SEC Chair Paul Atkins indicated last November that most cryptocurrencies, including meme coins, shouldn’t be treated as securities. That sentiment was bolstered by SEC guidance published last month, which categorized meme coins as a form of “digital collectibles.”

Under generic listing standards for crypto ETFs established last year, exchanges are able to list commodity-based ETFs without requiring case-by-case approval. Among key factors, digital assets underlying them have to have a six-month history of regulated futures trading.

Pepe futures currently trade on crypto exchange Kraken. Canary’s filing noted that contracts for the meme coin “are typically traded on regulated or registered trading venues.”

Canary has filed applications for ETFs that track other meme coins, including Mog, Pudgy Penguins’ PENGU, and President Donald Trump’s meme coin, TRUMP. Bloomberg Senior ETF Analyst Eric Balchunas expressed skepticism that the Trump-related ETF would pass when Canary’s application landed on the SEC’s desk last year, citing a lack of futures trading.

Balchunas once noted to Decrypt that the ETF industry is famous for “throwing spaghetti at the wall.” Meanwhile, Butterfill described a flurry of filings across ETFs from some issuers on Thursday as a “machine gun approach.”

Tuttle Capital Management, in some ways, has taken further steps to appeal to degens. In January, the ETF issuer filed applications for leveraged TRUMP, BONK, and MELANIA ETFs. But the SEC hasn’t offered a final verdict on those applications yet.

Enterprise revenue now makes up more than 40% of AI behemoth OpenAI’s total revenue, according to the company. And it’s on pace to reach parity with consumer revenue by the end of 2026.

OpenAI hit $25 billion in annualized revenue in February, up from $20 billion at the end of 2025.

“I have never seen this level of conviction spread so quickly and consistently within the industries,” OpenAI Chief Revenue Officer Denise Dresser, who spent more than a decade at Salesforce before running Slack, wrote in an official note on Wednesday.

Companies at the front of this wave have moved well past using AI to write emails or summarize documents. They’re now deploying what Dresser calls “teams of agents,” basically groups of AI systems that coordinate with each other, hold context across sessions, and take action inside business tools without constant human oversight. The question seems to have shifted from “should we use AI?” to “how many agents should we run?”

OpenAI launched its enterprise agent platform to build a user base beyond everyday retail consumers, who are still its core revenue stream. Codex, its AI coding agent, has already crossed 3 million users, a figure that was, according to Dresser, “almost zero” at the start of the quarter. Paying business users hit 9 million in February, up from 5 million in August. Weekly active users across all of OpenAI’s products reached 910 million.

The company also launched ChatGPT Agent, which can plan trips, book hotel rooms, research competitors, generate slide decks, and place online orders without a human in the loop.

But as hyped up as agentic AI is, Dresser believes companies need a straightforward path to integrate the tech without rebuilding their business structure.

“What’s really missing still for most companies is just a simple way to unleash the power of agents as teammates that can operate inside the business without the need to rework everything,” she said. OpenAI’s agent platform wants to be the answer to that problem.

OpenAI recently brought on Peter Steinberger, founder of the world’s most popular open source agentic AI platform OpenClaw, to lead its push into personal AI agents—a signal that the company isn’t only building for corporations. OpenAI CEO Sam Altman has positioned multi-agent systems at the center of OpenAI’s next product phase, and the momentum behind enterprise adoption suggests that framing is holding up in the market.

The company is also preparing for an IPO, with CFO Sarah Friar confirming this week that retail investors will get a share of the allocation. OpenAI projects reaching $85 billion in revenue by 2030—a number that only makes sense if agents become the default way businesses interact with AI, not just a feature layered on top of a chat interface.

Researchers from the Google Cloud AI team have introduced PaperOrchestra, an AI framework that autonomously transforms messy lab notes and scattered research data into submission-ready academic manuscripts.

Unlike existing AI writing tools that focus on text generation, the system aims to tackle the full intellectual workflow of academic paper creation—from organizing raw materials to generating figures and conducting literature reviews.

The system employs five specialized agents working in parallel: Outline Agent, Plotting Agent, Literature Review Agent, Section Writing Agent, and Content Refinement Agent. Each agent handles specific aspects of manuscript preparation, from structuring arguments to creating visualizations and ensuring proper academic citations through API-grounded references.

To evaluate performance, researchers created PaperWritingBench, the first standardized benchmark reverse-engineered from 200 top-tier AI conference papers. In side-by-side human evaluations, researchers noted, PaperOrchestra achieved win rate margins of 50%-68% for literature review quality and 14%-38% for overall manuscript quality compared to autonomous baselines.

PaperOrchestra emerges as AI systems are increasingly making inroads on knowledge work and specialized domains that are traditionally the preserve of humans, with the emergence of AI research agents and growing evidence of AI ghostwriting in academic papers.

The framework’s multi-agent approach—where specialized components tackle different aspects of a complex task—mirrors similar architectures being deployed across legal document analysis, financial modeling, and other domains requiring multi-step intellectual processes.

The use of AI tools in academic research has proved divisive, however, with some scholars dismissing the practice as “vibe coding,” and noting that the flood of AI-assisted papers in certain fields is putting “considerable strain” on peer-review systems.

Meta and CoreWeave have expanded a long-term artificial intelligence infrastructure agreement worth about $21 billion, extending their partnership through December 2032. Both companies have seen their shares rise Thursday following the announcement.

CoreWeave announced the expanded deal Thursday, describing it as an extension of the companies’ existing relationship and an increase in infrastructure supporting Meta’s AI operations.

“This is another example that leading companies are choosing CoreWeave’s AI cloud to run their most demanding workloads,” CoreWeave co-founder, CEO, and Chairman Michael Intrator said in a statement.

The agreement gives Meta access to AI cloud capacity from CoreWeave to support the development and deployment of its AI systems, including inference workloads that run trained models at scale.

The infrastructure will be deployed across multiple locations and will include some of the first deployments of Nvidia’s Vera Rubin platform. CoreWeave said the distributed deployment is intended to optimize performance, resilience, and scalability for Meta’s AI systems, and reflects rising demand for infrastructure capable of supporting large-scale AI workloads.

Meta and CoreWeave previously struck a $14 billion AI infrastructure deal in 2025, under which the cloud provider agreed to supply computing power to Meta through 2031.

The news comes as Meta accelerates its push into advanced AI systems. On Wednesday, the company introduced Muse Spark, a natively multimodal model capable of processing text, images, and voice and designed to tackle complex reasoning tasks using multiple AI agents.

Meta has also outlined a new Advanced AI Scaling Framework that expands how it evaluates risks and tests its most capable models before deployment.

“As we build more capable and more personalized AI, reliability, security, and user protections are more important than ever,” Meta said in a statement. “Advanced models require an advanced approach to safety—one that scales with the technology.”

Shares of Meta and CoreWeave rose after the companies announced their expanded AI infrastructure deal. Currently, Meta (META) is trading above $630 per share, up about 3% on the day, while CoreWeave (CRWV) has jumped 5.5% to a recent price of $93.70.