Bill Gates Foundation sells all $3.2 billion of its Microsoft holdings. Announces collaboration with Anthropic in the healthcare, education, and agriculture sectors.

The Bill & Melinda Gates Foundation Trust completely divested its Microsoft holdings in the first quarter of 2026, offloading its final remaining roughly 7.7 million shares (worth approximately $3.2 billion, or about ¥506 billion), according to a quarterly report (13F) disclosed on May 15. In nearly the same week, Anthropic and the Gates Foundation announced a four-year, $200 million (approximately ¥31.6 billion) collaboration targeting healthcare, education, and agriculture. Rather than treating these two news items as separate events, this article traces them as a single structural through-line that runs through Microsoft's move to "break free from sole dependence on OpenAI," Anthropic's surging corporate valuation, and all the way to "Mythos"—the frontier model that has become the tipping point of the cyber-offense-and-defense battle—across Silicon Va

Summary

It was revealed in a quarterly report (13F) disclosed on May 15 that the Bill & Melinda Gates Foundation Trust fully divested its Microsoft holdings in Q1 2026, offloading the final remaining roughly 7.7 million shares (approximately $3.2 billion, or around ¥506 billion). In nearly the same week, Anthropic and the Gates Foundation announced a four-year, $200 million (approximately ¥31.6 billion) collaboration targeting healthcare, education, and agriculture. Rather than treating these two news items as separate events, this article reads them from a Silicon Valley venture capital (VC) perspective as a single structural through-line connecting Microsoft's move away from "OpenAI as sole pillar," Anthropic's surging corporate valuation, and the frontier model "Mythos" — which has become a critical inflection point in the offense-defense dynamics of cybersecurity.

$3.2 Billion Total Sell-Off — Gates Foundation Trust Fully Exits Microsoft

On Friday, May 15, 2026, the Bill & Melinda Gates Foundation Trust's holdings report for the first quarter of 2026—the so-called 13F—filed with the U.S. Securities and Exchange Commission (SEC) was made public. What it revealed was the fact that the Trust no longer holds a single share of Microsoft stock. During the first quarter (January–March), the approximately 7.7 million shares that had remained until the very end were all sold off. These approximately 7.7 million shares correspond, at market prices, to roughly $3.2 billion (about ¥506 billion). Since the foundation's establishment in 2000, this is the first time the Trust has held no Microsoft stock at all.

This was not an abrupt sell-off, but rather the final chapter of a long, gradual withdrawal. According to tallies by Reuters and several financial media outlets, the Trust began trimming its holdings in late 2023, with the largest single reduction occurring in the third quarter of 2025, when about 65% of the position remaining at that point was disposed of. At the end of the first quarter of 2025, one year prior, it held approximately 28.5 million shares worth about $10.7 billion (approximately ¥1.69 trillion), accounting for roughly 26% of its portfolio. At the 2022 peak, Microsoft as a single stock made up an outsized share of about 27% of the entire Trust, so this quarter's reduction to "zero" marks the end of a quarter-century of concentration. After the withdrawal, the Trust's portfolio is reported to be on the order of about $31.7 billion (roughly ¥5 trillion).

Why sell? The answer is not "bearishness" but "mechanical inevitability." In May 2025, Bill Gates had announced a plan to wind down the foundation over 20 years. The plan calls for deploying approximately $200 billion (about ¥31.6 trillion) toward philanthropic activities by the end of 2045, donating 99% of Gates's personal remaining assets, and exhausting the endowment by that point. Having decided to "draw down" the endowment, the foundation needs abundant liquidity to distribute grants on the scale of tens of billions of dollars per year. Annual grant amounts are planned to be raised to about $9 billion (approximately ¥1.42 trillion) by 2026, and to keep disgorging that much cash, extreme concentration in a particular stock itself becomes a liquidity risk. In other words, a foundation entering its end-of-life phase "has no choice but to sell."

The market reaction was consistent with that interpretation. On May 15, 2026, when the 13F disclosure was reported, Microsoft stock actually closed up about 3.05% from the previous day at $421.92 (about ¥67,000). The rise that day itself appears to reflect separate factors that coincided in timing, such as U.S. President Trump's visit to China and U.S.–China talks, and is not directly linked to the foundation's sales. But at the very least, there was no development in which the discovery of the sale was disliked and caused the stock price to collapse—the market did not treat this disclosure as a "negative." The bulk of analysts as well positioned this not as a bearish signal regarding Microsoft's business outlook but as a portfolio adjustment driven by the foundation's structural and philanthropic circumstances. For a giant company with a market capitalization of approximately $3.13 trillion (about ¥495 trillion), the fact that a major shareholder of a quarter century had completely stepped down was no longer even an event that could move the stock price.

What was symbolic was that the exact opposite movement around the same stock was observed on the same day. Just before the 13F disclosure, Pershing Square Capital Management, led by prominent activist investor Bill Ackman, announced the acquisition of a new position in Microsoft. Pershing's 13F showed the acquisition of approximately 5.65 million shares, with a valuation at the end of the quarter (end of March) of about $2.09 billion (approximately ¥330 billion). Some media described this as a "bet" on the order of about $2.3 billion (approximately ¥363 billion). Ackman is said to have reduced his three-year-held Google position to zero while buying Microsoft at a forward P/E of about 21x, with the rationale reportedly being the view that "paid usage of Copilot is limited to only about 3% of addressable seats, and the franchise value of the enterprise is undervalued relative to AI uncertainty." A "philanthropic foundation stepping down" and a "prominent investor stepping up" moved in opposite directions on the same day regarding stock in the same company—this contrast, as will be discussed later, serves as foreshadowing for this article.

How the Gates Foundation Viewed Microsoft's Future — Four Ways to Read It

"How did the Gates Foundation view Microsoft's future?" This is the most divisive interpretive question raised by the recent reports of the share sale, and there is no single answer. At least four lines of reading are tenable.

The first is the official reading presented by the foundation itself: "This is not a verdict on Microsoft." As we saw in the previous chapter, the sale arose from a mechanical inevitability—reducing concentration risk and securing liquidity for the goal of "spending down" by 2045. For an institution that continues to disburse roughly $9 billion (about ¥1.42 trillion) in cash each year, systematically monetizing its most appreciated single holding is merely an operational judgment unrelated to Microsoft's future prospects. Several financial media outlets characterized the decision as "the trust behaved as a portfolio manager rather than as a sentimental founder." For the foundation's investment team, Microsoft is no longer a "stock of memories" but simply one massively swollen position.

The second is a more interpretive reading found among some market participants, who view the timing of the sale as suggestive. Microsoft's stock sits in a high range on the tailwind of the AI boom, and the company has substantially raised its fiscal year 2026 capital expenditure outlook to roughly $190 billion (about ¥30 trillion)—a figure that has put the market on edge about the recoverability of AI investment. It is rational, this view holds, for an institution that needs large amounts of cash on an annual basis to preferentially "harvest" the megacap stock that has appreciated the most and is most exposed to the burden of AI capex. This view often connects to the interpretation of "prudent profit-taking amid an overheated phase of AI valuations."

The third is a reading that relativizes this implicational argument. As noted in the previous chapter, the prominent investor Bill Ackman entered a new position in Microsoft stock during the same week, explaining that he bought at a forward P/E of around 21x—on par with the market average. His thesis is that paid penetration of Copilot is still in its early stages and that the enterprise franchise value is undervalued relative to the uncertainties of AI. In other words, the fact that "the foundation sold" does not signify a market consensus that "Microsoft is overvalued." The foundation was a seller driven by liquidity, not a seller who had given up on the future—that is the framing. In fact, Bill Gates personally is reportedly still holding a substantial amount of Microsoft stock, and what stepped back this time was strictly the "foundation trust" as an investment entity.

And the fourth—the reading this article weighs most heavily—is a VC-style reading of the sale as a "symbol." The foundation has not redirected the proceeds into AI stocks; rather, it is shifting its operational footing toward less volatile assets. This is a solid posture befitting an institution in its winding-down phase, bound by a grant schedule. Read the other way around, the foundation no longer feels the need to "hold the upside of AI through Microsoft stock." Instead, it is going after the social upside that AI brings not through equity ownership but through a "programmatic partnership"—with Anthropic, as we will see in the following chapters. To add an even greater irony, as we will examine in detail in a later chapter, Microsoft itself in 2026 is breaking free from its sole reliance on OpenAI and is incorporating Anthropic's Claude into its own Copilot. The very boundary between "Microsoft's future" and "Anthropic's future" is now dissolving. The foundation's sale of Microsoft shares is also a mirror reflecting that dissolution. What the foundation saw was not "Microsoft's decline" but a larger tectonic shift—"the era when one can bet on the future only through the vessel of Microsoft stock is over." That, it is fair to conclude, is the most adequate summary.

$200 Million Collaboration — Anthropic and the Gates Foundation Team Up on Healthcare, Education, and Agriculture

On May 14, 2026, the day before Bill Gates's complete divestment of Microsoft stock came to light, Anthropic and the Gates Foundation announced a four-year, $200 million (approximately ¥31.6 billion) collaboration. The first thing to grasp here is that this is not a "$200 million check in cash." According to both organizations' explanations, the $200 million consists of a combination of grant funding, Claude usage credits (API and usage allocations), and technical support. In terms of the division of roles, the Gates Foundation contributes grant funding, program development, and on-the-ground operational know-how, while Anthropic provides technical staff support and Claude usage allocations. It is positioned as "the largest arrangement of its kind" forged between an AI company and a global philanthropy.

The scope spans three domains. The largest pillar is healthcare and life sciences. Globally, approximately 4.6 billion people still lack access to essential medical services, with most of them concentrated in low- and middle-income countries. The collaboration aims to accelerate the development of vaccines and treatments and to render complex, massive medical data into forms that researchers and policymakers can handle with ease. Specifically, researchers will use Claude to conduct systematic reviews (comprehensive analyses of prior research) and detect patterns in large-scale datasets, and to computationally screen vaccine and drug candidates before they advance to preclinical stages. The initial priority diseases are polio, HPV (human papillomavirus, the leading cause of cervical cancer), and eclampsia and preeclampsia. HPV-related diseases bring about approximately 350,000 deaths annually, with around 90% of those concentrated in low- and middle-income countries. Through a partnership with the non-profit Institute for Disease Modeling, the collaboration will enhance the accuracy and usability of epidemiological forecasts that inform how malaria and tuberculosis treatment resources should be allocated across regions. On the technical side, Anthropic will develop "connectors" that link Claude directly to external health platforms — enabling national governments to leverage their own health data for policy decisions — and the scope also includes the development of benchmarks and evaluation frameworks for measuring the performance of medical AI. Notably, this healthcare collaboration is roughly four times the scale of "Horizon1000," the $50 million (approximately ¥7.9 billion) healthcare initiative that the Gates Foundation announced with OpenAI at Davos in January 2026 — a concept to start from Rwanda and expand to Kenya, South Africa, and Nigeria, delivering AI to 1,000 primary care clinics by 2028 — and is broader both geographically and across fields. The very fact that the Foundation is partnering with multiple frontier AI companies in parallel reflects a "public goods" orientation that avoids dependence on any single company.

The second domain is education. The focus here is on building the shared infrastructure needed to make AI function as a "tool for teaching and learning." If models can more accurately understand students' proficiency levels and "where they are getting stuck," teachers can detect students' difficulties earlier and deliver more individualized support. As a concrete framework, under the Global AI for Learning Alliance (GAILA), Claude will be applied in the United States to AI tutors for K-12 (elementary and secondary education), individualized math tutoring, college admissions advising, and curriculum design. In sub-Saharan Africa and India, AI applications will be developed to support foundational literacy and numeracy programs, and in the process, data collection and labeling (annotation) for African languages — which AI models find difficult to handle — will be advanced, with the results made public. And the development of public benchmarks, datasets, and knowledge graphs (data that structures knowledge) for verifying whether AI tutors are truly effective is expected to produce its first results as early as the latter half of 2026. Once these are released, the benefits will extend not just to Claude but to every AI system used in global health and education.

The third domain is agriculture, as referenced in the article's title. However, there is a subtlety in "how it is communicated," which will be discussed later. Worldwide, there are roughly 2 billion people whose livelihoods depend on smallholder farming, and many of them rely on limited information to make decisions about planting and responding to pests and diseases. The collaboration will refine Claude in a crop-specific manner and establish datasets and evaluation benchmarks for local crops as public goods, so that these farmers can use locally-relevant data in their own languages and make trustworthy real-time decisions. In parallel, primarily targeting the United States, efforts will advance to support people's "transition" to new jobs in an era when AI-driven employment automation is a concern — including portable records of skills and credentials that can move across organizations, career navigation for those entering the labor market, and mechanisms that tie vocational training data to actual employment outcomes.

At the core of the collaboration lies the concept of "public goods." Datasets, benchmarks, and infrastructure will be released in forms that anyone can freely use — so that progress in one region ripples out to others. Efforts to support data collection and labeling for African languages, lifting up model training across the entire industry, are part of this. Janet Zhou, a director at the Gates Foundation, explained the background behind this approach: "The policy of placing weight on public goods emerged from the needs of various partners and governments — including concerns about proprietary lock-in and sovereignty." Elizabeth Kelly, who leads the beneficial deployments team at Anthropic, stated, "This announcement goes to the core of who we are as a company."

It should be noted that this collaboration is not Anthropic's first move in the healthcare field. The company already announced "Claude for Life Sciences" for researchers and clinicians in October 2025, and at the JPMorgan Healthcare Conference in January 2026 announced "Claude for Healthcare" for healthcare institutions and payers. The latter is equipped with HIPAA (Health Insurance Portability and Accountability Act)-compliant infrastructure and integrates with foundational medical and scientific data such as the CMS coverage database, ICD-10 codes, and the literature database PubMed. The collaboration with the Gates Foundation is accurately understood as the "Global South / philanthropic edition" of this series of healthcare strategies.

Each Company's "Way of Telling It" — Three Narrative Styles: Charity, Mission, and Public Good

Even for the same $200 million collaboration, the narrative tone differs depending on who is doing the telling. And the differences themselves reflect where each organization stands.

The Gates Foundation's press release was titled "Making AI work for more people." The frame is the concrete development challenges the foundation has tackled for a quarter century: health, education, and agriculture. The protagonists are unwaveringly "people" and "challenges," with AI relegated to the background as a means. As Director Zhou's remarks indicate, the foundation foregrounded the language of "public goods," answering governments' concerns about sovereignty.

Anthropic's own blog, on the other hand, was titled "Anthropic forms $200 million partnership with the Gates Foundation," and described the target domains as "global health, life sciences, education, and economic mobility." Here lies the first subtle distinction. The domain the Gates Foundation names "agriculture" is bundled by Anthropic under the more abstract economic term "economic mobility." Within Anthropic's vocabulary, agriculture is part of "the economic range of motion through which people escape poverty." And Kelly's words about "the core of our company" tie the collaboration to the company's identity of championing safety and public benefit. For Anthropic, this collaboration is at once philanthropy and a demonstration of its brand as a "responsible AI company."

The media coverage was not uniform either. Many outlets led their headlines with scale: "$200 million partnership," "the largest of its kind." Some painted it as a problem-solving narrative—"filling AI's global blind spots"—while others, as discussed below, reported through the critical lens of "narrative immunity."

Here I want to point out an asymmetry of "delivery" that tends to be overlooked, and which is the most important of all. The Gates Foundation's "complete divestment of $3.2 billion in Microsoft stock" was not something the foundation actively "announced." It is a fact that, so to speak, unavoidably "came to light" through the institutional mechanism of quarterly-mandated 13F disclosures. The collaboration with Anthropic, by contrast, is an "announcement" carefully press-released with chosen dates and curated quotations. The former is institutional disclosure; the latter, a meticulously chosen narrative—when these two are placed side by side as "two simultaneous news stories," a narrative spontaneously emerges, as if the foundation had given up on Microsoft and switched over to Anthropic. Viewed coolly, however, one is the mechanical liquidation of assets by a foundation that has entered its wind-down phase, while the other is a programmatic partnership; there is no direct causal relationship between them. And yet—as the next chapter will discuss—the two resonate strongly at the symbolic level.

Why It Appears to Be "Happening Simultaneously" — The Tectonic Shift of Microsoft → OpenAI → Anthropic

There is no direct causal link between the two news items. Yet both rest on the same tectonic shift. The key to understanding that shift lies in Microsoft's own pivot in AI strategy.

Let me rewind history a bit. In 2019, when Microsoft was about to make its first $1 billion (roughly ¥158 billion) investment in OpenAI, co-founder Bill Gates warned CEO Satya Nadella that "you're going to burn that billion dollars"—an anecdote Nadella himself revealed publicly in 2026. Microsoft ultimately invested a cumulative total of roughly $13 billion (about ¥2 trillion) in OpenAI, and through OpenAI's restructuring in October 2025, it obtained about a 27% stake. That stake was reportedly valued at around $135 billion (about ¥21.3 trillion) at the time. How Gates's assessment of this investment ultimately turned out to be "correct" will be examined again in the next chapter.

However, Microsoft itself, entering 2026, began moving away from its "OpenAI-only" stance. Its flagship Microsoft 365 Copilot shifted to a multi-model configuration, and many enterprise inference tasks came to be handled by Anthropic's Claude. Claude is used for generation processes in Excel and PowerPoint, and Claude is also offered on Microsoft Foundry. In March, "Copilot Cowork" was launched in collaboration with Anthropic. Nadella explained this shift from a customer-centric perspective, saying "customers want to be able to choose from multiple models," and the media characterized it as "the end of the monopoly" and "the end of the single-model era." In short, part of Microsoft's most important software business is now running on the technology of Anthropic, OpenAI's rival.

Let us place this article's two news items within this configuration. First, the Gates Foundation completely exited Microsoft stock. Second, the Gates Foundation began an AI collaboration with Anthropic. The foundation's assets are no longer staked on Microsoft, and the AI partner the foundation has chosen to carry out its mission is not Microsoft (and OpenAI behind it) but Anthropic. To repeat, this is not a causal relationship. But it does constitute two independent pieces of evidence showing that the center of gravity of the "Gates world" has shifted from the company he co-founded in 1975 to the other pole of frontier AI. VCs (venture capitalists) read such "shifts in center of gravity" not as isolated news items but as structure.

"That $1 Billion Will Burn" — How Gates' Assessment of OpenAI Turned Out to Be Right

The remark Gates made to Nadella, mentioned in the previous chapter—"You're going to burn that $10 billion"—deserves a more careful pause as we work through the two news stories at the heart of this piece. To state the conclusion first: the story surrounding this investment can be summed up under the upbeat reading that "in the end, it turned out to be a brilliantly correct call."

To begin with, the soundness of the investment decision itself is no longer open to debate. On May 11, 2026, in the courtroom of the lawsuit Elon Musk brought against OpenAI and Microsoft, an internal memo that Microsoft President Brad Smith addressed to the board in January 2023 was introduced as evidence. It set out an ambitious figure: an expected return of roughly $92 billion (about ¥14.5 trillion) on a cumulative investment of about $13 billion (about ¥2 trillion) in OpenAI. And reality outstripped even that projection. As of early 2026, OpenAI is reported to be valued at approximately $852 billion (about ¥135 trillion), and the value of Microsoft's roughly 27% stake—though the figures vary by report—is said to have reached around $220–230 billion (about ¥35 trillion). That is fully 17–18 times the roughly $13 billion put in. Nadella, who testified for about two and a half hours in court, was asked about this memo and replied, "It worked out. Because we took the risk," while candidly acknowledging the precariousness of the bet by adding, "The return could just as easily have been zero." Testimony has also come to light that at the time of the investment, Nadella feared Microsoft would become "the next IBM." Nadella was the one who led the bet, and the numbers now prove that his judgment was a historically correct one.

So was Gates, who voiced skepticism, simply "wrong"? It is here that this piece wants to emphasize a slightly deeper reading. The heart of Gates's concern was not the size of the investment itself, but the fact that "OpenAI at the time was a nonprofit"—the danger of pouring a massive sum into a vessel that structurally promised no for-profit return. And on that single point, Gates's intuition was, if anything, on the mark. In November 2023, OpenAI's nonprofit board abruptly dismissed CEO Sam Altman, only to reinstate him within just a few days, throwing the organization into chaos. Then, in 2025, OpenAI at last restructured its corporate form into a for-profit entity (a Public Benefit Corporation). Gates's read—"a business on the scale of tens of billions of dollars cannot be sustained within this nonprofit vessel"—became reality, albeit in a different form. From the standpoint of investment returns, Nadella was right; from the standpoint of organizational sustainability, Gates was right—each, on his own ground, turned out to be "correct in the end."

What matters above all is that Gates, even while holding his skepticism, did not block the bet. His healthy skepticism worked not to kill the wager, but to make it more robust. In fact, Gates himself famously challenged OpenAI to "build me an AI that can pass a college-level (AP) biology exam," and that became a crucial milestone on the road to GPT-4. He has described the moment he first saw the GPT-4 demo as "stunning—it took my breath away," likening it to an experience on a par with seeing a graphical user interface for the first time in 1980. The initial skeptic was, in this way, transformed into one of AI's most eloquent advocates.

This story of "productive skepticism" is the other connecting line that ties together the two news items in this piece. Precisely because he witnessed the disruptive power of frontier AI more closely than almost anyone—and became convinced that its deployment is what will solve humanity's hardest problems—Gates's foundation is now, with its own hands, stepping into a bet on AI. This time not through an equity investment, but in the form of a partnership with Anthropic. Letting go of Microsoft stock and aligning closely with Anthropic are—even without a direct causal link between them—the front and back of one and the same conviction.

Silicon Valley VCs' Take — "Winner Concentration" and "Narrative Immunity"

From a VC perspective, both news stories serve as mirrors reflecting "the structure of the AI capital market in 2026."

First, there is the extreme concentration of capital. In February 2026, global funding flowing into startups reached approximately $189 billion (about ¥30 trillion), with roughly 90% of it—about $170 billion (about ¥27 trillion)—directed toward AI-related ventures. Capital is being funneled to a small number of "presumed winners." Anthropic itself raised approximately $30 billion (about ¥4.74 trillion) in a Series G round on February 12, 2026, reaching a post-money valuation of $380 billion (about ¥60 trillion). Led by GIC and Coatue, with joint participation from D.E. Shaw Ventures, Dragoneer, Founders Fund, ICONIQ, and MGX, this round ranks as the second-largest private fundraising in tech industry history. Funding for foundation model companies in Q1 2026 alone swelled to twice the total for all of 2025. VCs are no longer thinly diversifying across many model companies; they are concentrating capital on a single point.

On top of this, while there was once an unwritten rule that "you don't simultaneously invest in competitors," in 2026 a succession of VCs invested in both OpenAI and Anthropic. Given that the uncertainty of the AI race is too great to definitively call a winner, "betting on both" has become rational. Anthropic's corporate valuation appears to have no ceiling, and according to a Bloomberg report on May 12, 2026, the company is in discussions for a new round of approximately $900 billion (about ¥142 trillion). However, it should be noted that this is information only at the discussion stage and is not yet finalized.

Placed in this context, the Gates Foundation's sale of Microsoft can be read in VC terms as "a move away from old-generation tech," while its collaboration with Anthropic can be read as "a move toward the new frontier." That said, most VCs are coolly evaluating the foundation's sale, with the dominant view being, "That's a mechanical asset drawdown, not a bearish view on Microsoft." Ackman's contrarian buy backs up this perception.

VC assessments of the collaboration itself are broadly divided into two camps. The positive view sees this as a skillful go-to-market move by Anthropic. Healthcare, education, and agriculture in the Global South could all eventually become massive AI markets. The Gates Foundation's program network, government relationships, and on-the-ground trust are themselves the "distribution channel" needed to reach those markets. Even more important is the fact that a substantial portion of the $200 million consists of Claude usage credits and technical support. Anthropic is not so much donating cash as "distributing its own products and cultivating an ecosystem," which from a VC standpoint appears to be an extremely rational strategy.

Moreover, in the same week, Anthropic showed an entirely different face as well. On May 4, the company announced a joint venture of approximately $1.5 billion (about ¥237 billion) with Wall Street heavyweights such as Goldman Sachs and Blackstone. The plan envisions deploying Claude to private equity (PE) portfolio companies with embedded engineers, challenging the consulting industry. The fact that Anthropic reached out to both philanthropy (the Gates Foundation) and Wall Street (Goldman and Blackstone) in the same week is the clearest possible proof that the company is going after the market on all fronts.

On the other hand, a skeptical view also exists. Some commentators describe this $200 million as "a down payment on narrative immunity." Their logic goes like this: Anthropic needs to build a "permission structure" before job destruction and stronger regulation kick in in earnest, and the Gates Foundation, by lending its credibility from the world of development aid, neutralizes criticism in advance. In situations where journalists or policymakers might simply equate "AI = job destruction," Anthropic will be able to cite this collaboration as counter-evidence that "AI can be directed in good directions as well." The very fact that the collaboration's "public goods" orientation is designed as a response to governments' sovereignty concerns lends this interpretation a certain persuasiveness.

Viewed from a VC perspective, there is another structural shift that cannot be overlooked: the boundary between philanthropy and VC is dissolving. OpenAI is countering as well. The OpenAI Foundation has announced at least $1 billion (about ¥158 billion) in grants over the next year, which is part of a total $25 billion (about ¥3.95 trillion) philanthropic commitment targeting life sciences, employment and economic impact, AI resilience, and more. The OpenAI Foundation holds equity worth approximately $130 billion (about ¥20.5 trillion) and has already become the largest nonprofit organization in the United States. While the Gates Foundation heads toward winding down over 20 years, the "foundations" of AI companies are rising up at a scale rivaling or surpassing traditional philanthropic foundations—this is also a generational changing of the guard in philanthropic capital. The Gates Foundation–Anthropic collaboration sits precisely at the hinge of that changing of the guard.

If we limit our focus to the healthcare domain, the tailwind from private VC is also striking. According to Rock Health's tally, digital health startups raised approximately $4 billion (about ¥632 billion) in Q1 2026 alone, exceeding the same period a year earlier by about $1 billion. AI accounts for approximately 46% of total healthcare investment and approximately 60% of digital health fundraising. Bessemer Venture Partners, in its report "State of Health AI 2026," positions 2026 as "an inflection point," while Andreessen Horowitz (a16z) argues that "AI copilots in life sciences and medicine will scale skilled labor." The Gates Foundation–Anthropic collaboration can also be read as a complement filling the void of "public goods in low- and middle-income countries," a space that the massive funding flows of private VC have difficulty reaching.

What is Mythos — Frontier AI Reaches the Critical Point of Cyber Offense and Defense

Here, we enter the third and probably heaviest piece of reporting. The article's title asks, "What is happening with Mythos?"

On April 7, 2026, Anthropic unveiled the "Claude Mythos Preview." Mythos Preview is a general-purpose language model, but it demonstrates outstanding capability on computer security tasks. According to Anthropic's account, given a user's instructions, it can autonomously identify zero-day vulnerabilities (previously unknown vulnerabilities) in every major OS and every major browser, and even write working attack code (exploits) for them.

Its track record is concrete. Mythos discovered a bug that had survived for 27 years in OpenBSD, and a 16-year-old bug in the H.264 implementation of the video codec FFmpeg. Both are codebases that have been audited by human hands many times over. In Firefox, it discovered as many as 271 zero-day vulnerabilities, and against Firefox 147 it succeeded in producing working exploits 181 times. Compared with just two from the prior model Claude Opus 4.6, this is roughly a 90-fold leap. A browser JIT heap spray attack chaining four vulnerabilities, remote code execution against the FreeBSD NFS server (obtaining root privileges without authentication), Linux kernel privilege escalation, and more — Mythos demonstrated each of these autonomously. The vulnerabilities discovered are said to number in the thousands for high- and critical-severity alone, and in the tens of thousands across software as a whole, and Anthropic itself acknowledges that over 99% of the vulnerabilities it discovered remain unpatched.

The cost structure was also shocking. Scanning 1,000 code entry points in OpenBSD reportedly cost less than $20,000 (about ¥3.16 million), individual exploit development ran $1,000–$2,000 (about ¥160,000–¥320,000) in API fees, and required only a few hours to about a day. Mythos made visible the fact that cyberattack capabilities, which until now required advanced specialist skill and commensurate cost, are becoming dramatically cheaper, faster, and within anyone's reach.

It is precisely for this reason that Anthropic explicitly stated, "We will not make Mythos Preview generally available." The reason for not releasing it publicly is the cyber danger that the capability itself brings. In its place, the company launched "Project Glasswing." This is an industry consortium aimed at protecting the world's critical software before attackers gain equivalent capability. The launch partners are 12 entities: AWS, Apple, Microsoft, Google, NVIDIA, Cisco, Broadcom, CrowdStrike, Palo Alto Networks, JPMorganChase, the Linux Foundation, and Anthropic. Access has further been extended to more than 40 organizations responsible for critical infrastructure, with roughly 50 organizations participating in total. Anthropic has committed up to $100 million (about ¥15.8 billion) in usage credits, and approximately $4 million (about ¥630 million) in donations to open-source security organizations.

Anthropic's CEO Dario Amodei called this situation a "moment of danger." "This is a moment of danger. If we respond correctly — and I think we have taken the first step — there is a better world beyond it." He also said, "The danger is that the number of vulnerabilities, the number of breaches, and the financial damage from ransomware against schools and hospitals, to say nothing of banks, will increase by orders of magnitude." These remarks were made at an Anthropic financial-services event in early May, where he appeared on stage alongside JPMorgan Chase CEO Jamie Dimon. Fortune reports that at this event, both CEOs sidestepped the question of whether this cyber "freakout" is warranted.

What Happens in Mythos — A Multifaceted Perspective

Views on Mythos are sharply divided among experts, regulators, and geopolitical actors. Since this lies at the heart of what the article's title asks, I want to introduce as many angles as possible.

First, there is the clash between the "overreaction" thesis and the "real threat" thesis. Financial regulators clearly moved in haste. In early April, Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent hastily convened a meeting in Washington with the CEOs of major banks gathered there—Bank of America, Citi, Goldman Sachs, Morgan Stanley, Wells Fargo, and others—to discuss the cyber risks raised by Mythos (JPMorgan CEO Jamie Dimon was reportedly unable to attend the meeting). The European Central Bank (ECB) called on eurozone banks to "urgently prepare for AI-powered attacks," the Bank of England's Cross Market Operational Resilience Group and AI Task Force have discussions planned, and the U.S. House Homeland Security Committee also received a briefing. But multiple cybersecurity experts and AI researchers interviewed by CNBC on May 8 characterized these movements as "hysteria." Their argument is clear—many of the vulnerabilities Mythos exposed can also be found using existing models from Anthropic or OpenAI. What is decisively new is the "next step," namely the autonomy to even craft exploits that operate with almost no human intervention, but criminal organizations and hostile-state hackers already possess that skill set. Ben Harris, CEO of cybersecurity firm watchTowr, stated, "The industry is now learning that results very similar to Mythos can be reproduced simply by skillfully combining publicly available models." The "real problem" experts point to is that while AI accelerates the discovery of vulnerabilities, it still takes companies days to weeks to apply patches (fixes), and this "discovery-to-fix gap" continues to widen. AI-driven cyberattacks increased by 89% in 2025, and the grace period before vulnerabilities are exploited has been dramatically shortened from 771 days in 2018 to just 4 hours in 2024. From this perspective, Mythos has merely made visible "a situation that was already underway."

Second, there is the debate over "inequality of access" and "sovereignty." Access to Mythos is limited to about 50 organizations, leaving many of the world's central banks and governments out of the loop. The White House reportedly rejected a proposal to expand access to about 70 organizations. This creates an "inequality of defensive capability." Chandramouli Dorai of Zoho criticized that "the organizations that need the most protection are excluded. Security must not be a luxury," and Nick Srnicek of King's College London pointed out that countries relying on their own digital sovereignty solutions rather than widely patched U.S.-made software would actually be at a disadvantage. An analysis on the legal commentary site Just Security frames as problematic that European allies were left "out of the loop" on Mythos access decisions, that a single private U.S. company is effectively making decisions that touch on national security, and that access restrictions actually deepen allies' dependence on the United States. The Next Web described this situation as "Mythos is traveling between governments faster than regulators can keep up"—institutions are failing to keep pace with technology. What comes to mind here is the fact that the collaboration with the Gates Foundation foregrounded "public goods" and "consideration for sovereignty." Even the most well-intentioned AI philanthropy cannot escape carrying the politics of dependence on U.S. AI companies. Mythos confronts us with that same dependency structure in its most starkly exposed form.

Third, there is the phrasing "AI bugocalypse." Prominent security expert Alex Stamos said, "Every kind of company, including your local dental clinic, will be forced to deal with new bugs. This is the AI bugocalypse." In April 2026, the World Economic Forum (WEF) called this the "Mythos Moment," arguing that frontier AI is redefining cybersecurity itself.

Fourth, there is the cold-eyed view that "leakage and proliferation are inevitable." In fact, Mythos was already accessed by an unauthorized group in April. An anonymous "private online forum" infiltrated via a third-party vendor's environment. The method is described as "low altitude, high impact"—they inferred URL patterns from Anthropic's naming conventions, used metadata obtained from a past data leak at AI startup Mercor, and exploited a contractor's credentials. Anthropic stated that "no impact on its own systems has been confirmed," and the group reportedly avoided detection by not submitting cyber-related prompts to the model and instead having it perform only simple tasks such as building websites. This is a real-world example showing that even the most tightly guarded models can leak. Note that China has been completely excluded from Project Glasswing. Chinese companies are cut off from direct access to Mythos and have no choice but to rely on independent development. However, the U.S. government claims that China is attempting to steal American AI through an "industrial-scale distillation campaign," using tens of thousands of proxy accounts to evade detection. Amodei himself has warned that there is only a "6–12 month" window before Chinese AI acquires Mythos-class capabilities, and researchers at cloud security firm Wiz have offered similar assessments.

Fifth, there is the view that "this is just the beginning." The Just Security analysis literally positions Mythos as a "preview" and warns that governments and companies must establish a comprehensive strategy before more powerful models emerge. Competition is also intensifying. Weeks after the Mythos announcement, OpenAI CEO Sam Altman unveiled "GPT-5.5-Cyber," specialized for cybersecurity, opening limited access to cybersecurity teams that pass vetting. According to CNBC's May 11 report, OpenAI intends to give the EU access to this model, while Anthropic still restricts the provision of Mythos. Offense and defense, disclosure and concealment, competition and cooperation—Mythos is simultaneously amplifying all of these tensions.

When and what will be measured going forward — reading the next moves on the timeline

Regarding the question "when and what kind of new developments will be observed going forward," I want to organize multiple observation points of differing confidence levels along a timeline.

In the immediate term, the first milestone will arrive by the summer of 2026. The most certain of these is Project Glasswing's "90-day report." Anthropic has promised that within 90 days of the project's launch (April 7–9), it will publish what it has learned, the vulnerabilities that have been fixed, and improvements within the disclosable scope. Working backward, early July 2026 will be the first externally verifiable "report card." For the first time, it will become possible to measure how many of the "tens of thousands of unpatched vulnerabilities" have actually been eliminated. In parallel, the safety mechanisms to be embedded in the next Claude Opus warrant attention. Anthropic has indicated a plan to refine new safeguards (mechanisms to detect and block the most dangerous outputs) on models that do not carry Mythos-class danger, and then deploy them in the next Claude Opus. This is a prerequisite for "safely deploying a Mythos-class model at scale someday." On the financial side, another observation point is whether the funding talks of roughly $900 billion (approximately ¥142 trillion) reported by Bloomberg on May 12 will be finalized over the summer. If finalized, Anthropic's "war chest"—simultaneously advancing philanthropy (the Gates Foundation), Wall Street (Goldman, Blackstone), and global expansion—will be further bolstered.

In the medium term, the latter half of 2026 lines up more geopolitical observation points. The biggest focus is whether the "6 to 12 month window" Amodei speaks of will close. From late 2026 into the first half of 2027, the question is whether Chinese models or high-performance open models will reach Mythos-class capabilities—if they do, the fundamental premise of Glasswing, that "the defender protects first," collapses. On the diplomatic front, the trajectory of US-China "guardrail" talks becomes an observation point. President Trump returned on May 15 from the first US presidential visit to China since 2017, telling reporters that he had discussed with President Xi Jinping the "possibility of cooperating on AI guardrails." According to senior White House officials, "frontier AI models that expose vulnerabilities in national cyber infrastructure" was one of the key agenda items in Beijing, and Mythos has already been elevated to a topic of summit-level diplomacy. Going forward, the question is whether bilateral or multilateral AI cyber agreements will be concluded. On the regulatory side, voices such as Just Security argue that "voluntary corporate safeguards are insufficient, and consistent, verifiable safety measures should be imposed on all models through regulation and negotiation," and the focus is whether Mythos will trigger the institutionalization of "pre-deployment AI model inspection and oversight" in the US and EU.

That said, the differences in confidence levels need to be made explicit here. Project Glasswing's 90-day report is Anthropic's official commitment and carries the highest confidence. By contrast, the $900 billion-scale funding is information only at the "negotiation stage," the timing of China's catch-up is ultimately Amodei's personal assessment, and the timing of regulatory institutionalization is even more uncertain. Rather than pinning "when" to a single point, it is appropriate to view it with a range.

Overall, the major observation points for the next 12 months can be consolidated into six. First, the substance of the Glasswing 90-day report, expected in early July. Second, the effectiveness of the safety mechanisms installed in the next Claude Opus. Third, the success or failure of Anthropic's $900 billion-scale funding. Fourth, the timing at which Chinese models reach Mythos-class. Fifth, the progress of US-China AI guardrail talks. And sixth, the first concrete deliverables born from the collaboration with the Gates Foundation—public benchmarks and datasets, drug candidates for HPV and preeclampsia. When any of these six moves, it will be the first answer from reality to the question, "what will happen with Mythos."

Conclusion — The Single Thread Running Through the "Three Reports" from a VC's Perspective

On the surface, this paper has dealt with three separate events: the Gates Foundation's complete divestment of its Microsoft shares, the collaboration between Anthropic and the Gates Foundation, and the frontier model Mythos. Yet when integrated through the lens of a Silicon Valley VC, a single thread running through all three comes into view.

That thread is the fact that three centers of gravity surrounding AI—"capital," "legitimacy," and "capability"—are shifting simultaneously. The Gates Foundation's sale of Microsoft shares is a story about "capital"—funds that had been staked on the previous generation of tech are leaving the hands of a philanthropic foundation winding down its affairs, transformed into liquidity for the frontier AI era. The collaboration with Anthropic is a story about "legitimacy"—a movement in which AI companies seek to procure social trust through the most public-facing domains: healthcare, education, and agriculture. And Mythos is a story about "capability"—unmistakable evidence that this capability has reached a dangerous level requiring control, no longer something that can be fully captured by philanthropic PR alone.

Herein lies the cruelly clear duality revealed by this single week in May 2026. The same AI company, Anthropic, presented to the world in roughly the same week both the most hopeful narrative—"using AI for good in healthcare, education, and agriculture" (the collaboration with the Gates Foundation)—and the darkest narrative—"AI could weaponize software around the world within hours" (Mythos). This coexistence of light and shadow is nothing less than the essence of frontier AI in 2026.

The implications for VCs are weighty. Investment judgments on AI can no longer be closed out with just "growth rate and market size (TAM)." First, winner concentration—capital tilting extremely toward a small number of companies. Second, the dissolution of the boundary between philanthropy and VC—AI company foundations beginning to surpass traditional philanthropic foundations in scale. Third, the externality of capability—as with Mythos, the model's capability itself transforming into a national security issue. VCs have entered an era in which these three must be simultaneously woven into investment decisions.

In the very same week that the foundation of Bill Gates quietly let go of shares in the company he co-founded in 1975, Gates's foundation joined hands with the AI company that symbolizes the next era, and another model from that AI company shook the world's cyber order. A quarter-century ago, Gates defined an era with Microsoft. This single week in May 2026 may be recorded as the moment when the "capital page" of that era was turned silently, and on the next page the headlines "Anthropic" and "Mythos" were inscribed. What happens next—the answer should first be revealed by the six observation points raised in this paper, and above all by Glasswing's 90-day report in July.

Sources

Anthropic "Anthropic forms $200 million partnership with the Gates Foundation" — https://www.anthropic.com/news/gates-foundation-partnership
Gates Foundation press release "Making AI work for more people" — https://www.gatesfoundation.org/ideas/media-center/press-releases/2026/05/ai-anthropic-partnership
The Next Web "Anthropic commits $200M with Gates Foundation to deploy AI in global health, education, and agriculture" — https://thenextweb.com/news/anthropic-gates-foundation-ai-health-education-partnership
Fortune "Gates Foundation, OpenAI unveil $50 million 'Horizon1000' initiative to boost health care in Africa" — https://fortune.com/2026/01/21/gates-foundation-openai-50-million-partnership-africa-rwanda-horizon1000/
TechStartups "Anthropic and Gates Foundation launch $200M partnership" — https://techstartups.com/2026/05/14/anthropic-and-gates-foundation-launch-200m-partnership-to-bring-ai-to-healthcare-education-and-underserved-communities/
U.S. News / Reuters "Anthropic, Gates Foundation Launch $200 Million Partnership for AI in Health, Education" — https://money.usnews.com/investing/news/articles/2026-05-14/anthropic-gates-foundation-launch-200-million-partnership-for-ai-in-health-education
The Information "Gates Foundation Sells Remaining Microsoft Stake" — https://www.theinformation.com/briefings/gates-foundation-sells-remaining-microsoft-stake
24/7 Wall St. "Bill Gates Just Did the Unthinkable — He Sold Every Last Share of Microsoft Stock" — https://247wallst.com/investing/2026/05/16/bill-gates-just-did-the-unthinkable-he-sold-every-last-share-of-microsoft-stock/
CoinCentral "Gates Foundation Sells Last of Its $3.2B Stake" — https://coincentral.com/microsoft-msft-stock-gates-foundation-sells-last-of-its-3-2b-stake-heres-why/
Parameter "Gates Foundation Trust Completes $3.2B Microsoft Exit in Q1 2026" — https://parameter.io/gates-foundation-trust-completes-3-2b-microsoft-msft-exit-in-q1-2026/
Moneywise "Bill Gates' foundation sold every Microsoft share — Bill Ackman is loading up" — https://moneywise.com/news/investing/bill-gates-foundation-microsoft-stock-bill-ackman-pershing-square-capital
Macrotrends "Microsoft (MSFT) Stock Price History" (May 15, 2026 closing price $421.92, up 3.05% from the previous day) — https://www.macrotrends.net/stocks/charts/MSFT/microsoft/stock-price-history
CNN Politics "Trump's China state visit and meetings with Xi Jinping (May 13–15, 2026)" — https://www.cnn.com/politics/live-news/trump-china-visit-xi-meeting-hnk
The Hill "Trump says he discussed AI guardrails with Xi" — https://thehill.com/homenews/administration/5880013-donald-trump-xi-jinping-china-summit-ai-guardrails/
Gates Foundation "Gates Foundation Will Double Spending Over Next 20 Years" (May 2025) — https://www.gatesfoundation.org/ideas/media-center/press-releases/2025/05/25th-anniversary-announcement
CNBC "Bill Gates to close foundation, give away rest of wealth by 2045" — https://www.cnbc.com/2025/05/08/bill-gates-to-close-foundation-give-away-rest-of-wealth-by-2045.html
Anthropic "Claude Mythos Preview" (red.anthropic.com) — https://red.anthropic.com/2026/mythos-preview/
Anthropic "Project Glasswing" — https://www.anthropic.com/glasswing
NBC News "Anthropic Project Glasswing: Mythos Preview gets limited release" — https://www.nbcnews.com/tech/security/anthropic-project-glasswing-mythos-preview-claude-gets-limited-release-rcna267234
CNBC "Anthropic CEO warns of cyber 'moment of danger' as AI exposes thousands of vulnerabilities" — https://www.cnbc.com/2026/05/05/anthropic-ceo-cyber-moment-of-danger-mythos-vulnerabilities.html
CNBC "Anthropic's Mythos set off a cybersecurity 'hysteria.' Experts say the threat was already here" — https://www.cnbc.com/2026/05/08/anthropic-mythos-ai-cybersecurity-banks.html
CNBC "Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks" — https://www.cnbc.com/2026/04/10/powell-bessent-us-bank-ceos-anthropic-mythos-ai-cyber.html
CNBC "OpenAI to give EU access to new cyber model but Anthropic still holding out on Mythos" — https://www.cnbc.com/2026/05/11/openai-eu-cyber-model-anthropic-mythos-gpt.html
Fortune "Jamie Dimon and Dario Amodei sidestep question about whether the AI cyber 'freakout' is warranted" — https://fortune.com/2026/05/05/jamie-dimon-dario-amodei-jpmorgan-anthropic-ai-cyber-mythos-freakout-sidestep-question/
Rest of World "Anthropic's Mythos and the global cybersecurity gap" — https://restofworld.org/2026/ai-cybersecurity-anthropic-mythos/
World Economic Forum "Anthropic's Mythos moment: how frontier AI is redefining cybersecurity" — https://www.weforum.org/stories/2026/04/anthropic-mythos-ai-cybersecurity/
Just Security "Too Dangerous to Deploy: Anthropic's Mythos and What Comes Next" — https://www.justsecurity.org/138011/too-dangerous-anthropic-mythos/
TechCrunch "Unauthorized group has gained access to Anthropic's exclusive cyber tool Mythos" — https://techcrunch.com/2026/04/21/unauthorized-group-has-gained-access-to-anthropics-exclusive-cyber-tool-mythos-report-claims/
CSO Online "What happens when China's AI catches up to Mythos?" — https://www.csoonline.com/article/4170818/what-happens-when-chinas-ai-catches-up-to-mythos.html
TechTimes "Trump and Xi Signal AI Safety Protocol Talks as Anthropic's Mythos Reframes the Threat" — https://www.techtimes.com/articles/316732/20260516/trump-xi-signal-ai-safety-protocol-talks-anthropics-mythos-reframes-threat.htm
Nextgov/FCW "House Homeland panel gets briefing on Anthropic's Mythos" — https://www.nextgov.com/artificial-intelligence/2026/05/house-homeland-panel-gets-briefing-anthropics-mythos/
NL Times "ECB warns banks about cyberattacks using Anthropic's Mythos AI model" — https://nltimes.nl/2026/05/13/ecb-warns-banks-cyberattacks-using-antrophics-mythos-ai-model
Anthropic "Anthropic raises $30 billion in Series G funding at $380 billion post-money valuation" — https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation
CNBC "Anthropic closes $30 billion funding round at $380 billion valuation" — https://www.cnbc.com/2026/02/12/anthropic-closes-30-billion-funding-round-at-380-billion-valuation.html
Bloomberg "Anthropic in Talks to Raise $30 Billion at $900 Billion Valuation" — https://www.bloomberg.com/news/articles/2026-05-12/anthropic-in-talks-to-raise-30-billion-at-900-billion-valuation
Bloomberg "VCs Break Taboo by Backing Both Anthropic, OpenAI in AI Battle" — https://www.bloomberg.com/news/articles/2026-02-11/vcs-rush-to-back-rivals-openai-and-anthropic-as-ai-frenzy-mounts
TechCrunch "Just three companies dominated the $189B in VC investments last month" — https://techcrunch.com/2026/03/03/openai-anthropic-waymo-dominated-189-billion-vc-investments-february-crunchbase-report/
CNBC "Anthropic teams with Goldman, Blackstone and others on $1.5 billion AI venture" — https://www.cnbc.com/2026/05/04/anthropic-goldman-blackstone-ai-venture.html
The Information "Microsoft to Buy AI From Anthropic in Partial Shift From OpenAI" — https://www.theinformation.com/articles/microsoft-buy-ai-anthropic-shift-openai
GeekWire "Microsoft 365 Copilot and the end of the single-model era in enterprise AI" — https://www.geekwire.com/2026/microsoft-365-copilot-and-the-end-of-the-single-model-era-in-enterprise-ai/
Fortune "Microsoft CEO Satya Nadella says Bill Gates told him his big bet on OpenAI would be a flop" — https://fortune.com/article/microsoft-ceo-satya-nadella-bill-gates-openai-sam-altman-youre-going-to-burn-this-billion-dollars-big-tech/
Bloomberg "Microsoft Targeted $92 Billion Return on Early OpenAI Investment" — https://www.bloomberg.com/news/articles/2026-05-11/microsoft-targeted-92-billion-return-on-early-openai-investment
GeekWire "Musk v. Altman: Satya Nadella was worried about Microsoft being 'the next IBM' in OpenAI deal" — https://www.geekwire.com/2026/musk-v-altman-satya-nadella-was-worried-about-microsoft-being-the-next-ibm-in-openai-deal/
Gates Notes "The Age of AI Has Begun" (Gates's review of GPT-4 and the AP Biology exam anecdote) — https://www.gatesnotes.com/The-Age-of-AI-Has-Begun
CNBC "Microsoft calls for $190 billion in 2026 capital spending on soaring memory prices" — https://www.cnbc.com/2026/04/29/microsoft-msft-q3-earnings-report-2026.html
Anthropic "Advancing Claude in healthcare and the life sciences" — https://www.anthropic.com/news/healthcare-life-sciences
Fierce Healthcare "JPM26: Anthropic launches Claude for Healthcare" — https://www.fiercehealthcare.com/ai-and-machine-learning/jpm26-anthropic-launches-claude-healthcare-targeting-health-systems-payers
Fierce Healthcare "Digital health startups raked in $4B during Q1" (Rock Health) — https://www.fiercehealthcare.com/digital-health/digital-health-startups-raked-4b-q1-12-megadeals-driving-investment-rock-health
Bessemer Venture Partners "State of Health AI 2026" — https://www.bvp.com/atlas/state-of-health-ai-2026
Andreessen Horowitz "AI at the Intersection: The a16z Investment Thesis on AI in Bio + Health" — https://a16z.com/ai-at-the-intersection-the-a16z-investment-thesis-on-ai-in-bio-health/
OpenAI "Built to benefit everyone" — https://openai.com/index/built-to-benefit-everyone/
The Chronicle of Philanthropy "Why the $130 Billion OpenAI Foundation Has Other Nonprofits on Edge" — https://www.philanthropy.com/news/why-the-130-billion-openai-foundation-has-other-nonprofits-on-edge/