Anthropic S-1 filing day was Monday 26 October 2026, when the maker of Claude made its initial public offering paperwork public with the US Securities and Exchange Commission. Most readers went straight to the money: revenue above $16 billion in the first half of 2026, a net loss of $80.7 billion, and a planned Nasdaq listing under the ticker ANT. But the largest section of the document is not about money at all. It is the risk factors, and they run from page 26 to page 119.
WIRED’s Brian Barrett put it neatly: the Anthropic S-1 is also an AI doomsday primer. Read in order, those 94 pages explain how a frontier artificial intelligence company thinks things could go badly wrong, from models that help build their successors to agents that act without anyone watching. The same document then sells one of those mechanisms to investors as a core advantage.
This article works through the Anthropic S-1 as a primer. We set out each doomsday mechanism the filing names, quote the exact language with page numbers, and show what the company says it has already done about each one. We also count how often the filing describes the same process as a risk and how often as a selling point, and explain what it all means for businesses that use Claude.
Table of contents
- What the Anthropic S-1 Is and Why It Reads Like a Primer
- Mechanism One in the Anthropic S-1: Closing the Loop
- Mechanism Two: Autonomy at Scale
- Mechanism Three: Strategic and Self-Preserving Behaviour
- Mechanism Four: Weapons, Cyberattacks and Escapes
- Mechanism Five: AI That Leaves the Screen
- What the Anthropic S-1 Says Has Already Been Done
- The Anthropic S-1 Counted: Advantage Versus Risk
- Why Put a Doomsday Primer in an IPO Filing?
- The Outside Pressure Named in the Anthropic S-1
- What the Anthropic S-1 Means for Businesses Using Claude
- What Happens Next to the Anthropic S-1
- Anthropic S-1 FAQs
- References
What the Anthropic S-1 Is and Why It Reads Like a Primer
A Form S-1 is the registration statement a company files before selling shares to the public. Its job is to tell investors everything material, and US securities law punishes companies that leave out risks they knew about. That legal pressure is why risk factor sections are long. It is also why this one reads like a field guide.
Six confidential drafts, then a public filing
The Anthropic S-1 did not arrive out of nowhere. EDGAR, the SEC’s filing system, now shows six confidential submissions before the public document. The table lists them with their filing dates.
| Date (2026) | Filing | What it was |
|---|---|---|
| 1 June | DRS | First confidential draft registration statement |
| 13 July | DRS/A | Confidential amendment |
| 10 August | DRS/A | Confidential amendment |
| 4 September | DRS/A | Confidential amendment |
| 1 October | DRS/A | Confidential amendment |
| 9 October | DRS/A | Confidential amendment |
| 26 October | S-1 | Public registration statement |
From the first draft on 1 June to the public filing on 26 October is 147 days. In late September, Reuters described a draft it had seen, and we covered that report in our piece on what the Anthropic prospectus showed about losses and growth. The public Anthropic S-1 is the first version anyone outside the process can read in full.
94 pages of risk
The table of contents puts “Risk Factors” at page 26 and the next section at page 120, so the risk section fills 94 pages. The main body of the Anthropic S-1 runs to about 335 numbered pages before the financial statements, which makes risk roughly 94 ÷ 335 ≈ 28 percent of it.
Reuters said the draft it saw had about 80 risk pages out of 261. On that basis the risk section grew by 14 pages, or about 17.5 percent, between that draft and the public Anthropic S-1. Some of the new material is plainly recent: the filing mentions events from September and October 2026.
The numbers investors will read first
The doomsday material sits beside some of the fastest revenue growth ever disclosed in an IPO filing. The table pulls the key figures from the summary financial data in the Anthropic S-1.
| Measure | Figure | Period |
|---|---|---|
| Revenue | $16,286 million | Six months to 30 June 2026 |
| Revenue | $4,585 million | Full year 2025 |
| Revenue growth | 1,300% | H1 2026 vs H1 2025 ($1,163 million) |
| Operating loss | $693 million | Six months to 30 June 2026 |
| Net loss | $80,748 million | Six months to 30 June 2026 |
| Fair-value charge on notes, derivatives and SAFEs | $72,388 million | Six months to 30 June 2026 |
| Customers spending $100,000+ a year | More than 11,900 (from 950+) | 30 September 2026 vs a year earlier |
The fair-value charge is about 72,388 ÷ 80,748 ≈ 90 percent of the net loss. It is an accounting effect of the company’s soaring valuation on its convertible instruments, not cash spent. The operating loss is much smaller. The offering price, share count and amount raised are still blank in the document.
The chart shows how fast revenue has moved. Each bar is a share of the first-half 2026 figure: for example, 2025’s $4,585 million ÷ $16,286 million ≈ 28 percent.
Mechanism One in the Anthropic S-1: Closing the Loop
The first doomsday mechanism is the one WIRED built its piece around. Researchers call it recursive self-improvement: AI systems doing the research and development that produces the next, stronger AI system. The worry is that the cycle speeds up until humans can no longer check what comes out of it.
The risk version
The key passage sits on page 47 of the Anthropic S-1. It says concerns have been raised, “both within our organization and by other researchers, regulators, policymakers, and industry participants”, that “sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity”.
The next sentence ties that outcome to a process. “We have observed that our models are contributing to the pace of our own research,” the filing says, “and these risks will increase as AI systems substantially automate AI research and development if, for example, they accelerate AI progress beyond our ability to evaluate and safeguard new models.”
Page 29 makes the same point in business language. The company’s “increasing use of AI systems to accelerate our own research and development may cause the capabilities of our models to advance more rapidly than our ability to develop, validate, and implement corresponding safety, alignment, and security measures.”
The advantage version
Elsewhere, the Anthropic S-1 describes exactly the same process as a strength. “Each version of a Claude model assists in building the next,” the summary says on page 6. “We refer to this recursive dynamic as ‘closing the loop,’ and we believe it is one of the most significant structural advantages in frontier AI.”
The business section supplies hard numbers. On page 184, the filing says over 90 percent of the contributions to Anthropic’s codebase in September 2026 were written by Claude. The average daily volume of code contributed per engineer in the three months to 30 September 2026 was over 45 times the 2024 average. The value of Anthropic’s own internal Claude usage at list prices grew more than 250 times between January and September 2026. In April 2026 that monthly value first exceeded the company’s monthly personnel expense, excluding stock-based compensation.
The same page says Claude’s success rate on “more challenging, open-ended engineering problems” that “historically depended on experienced human judgment” reached about 90 percent in September 2026. That was up about 50 percentage points in nine months on Anthropic’s internal benchmarks.
The phrase the filing keeps at arm’s length
The words “recursive self-improvement” appear once in the Anthropic S-1, on page 185, inside quotation marks. “Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress,” the filing says. The Responsible Scaling Policy, it adds, “sets out our plans to monitor and mitigate the risks as we approach what has been termed ‘recursive self-improvement.'”
“What has been termed” is careful drafting. The company names the mechanism, attributes the label to others, and says it is approaching it. Our earlier explainer on Anthropic’s own recursive self-improvement warning found the same reluctance in its research essay, which never used the acronym RSI at all.
Falling behind is a risk too
The Anthropic S-1 also lists the opposite danger. If Anthropic fails to direct enough compute and research to AI-assisted development, page 29 says, “we risk falling progressively further behind as competitors continuously use AI to further enhance their own development capabilities.”
So the filing names two risks that pull in opposite directions. Closing the loop too fast could outrun safety. Closing it too slowly could cost the company its place at the frontier. WIRED’s summary of that tension was blunt: the systems causing the existential concern are the ones the company sees as critical to its success.
Mechanism Two: Autonomy at Scale
The second mechanism is autonomy. A chatbot that answers questions can only do so much harm. An agent that holds real permissions inside a company, and works for days without a person checking each step, is different.
Agents with enterprise permissions
Page 47 of the Anthropic S-1 describes AI agents that are “capable of autonomously taking actions on behalf of users, interacting directly with external systems, executing code, making API calls, sending communications, or completing transactions with limited human oversight”.
These systems, it says, “are designed to possess enterprise-level permissions, integrations, and contextual awareness and to operate autonomously for extended durations, including sustained multi-step workflows lasting hours or days”. The filing warns that errors or misalignment may cause consequences “that may be difficult or impossible to reverse, such as committing code changes, deleting or modifying data, or completing financial transactions—before effective intervention is possible.”
Harm at scale without human intervention
The second “existential risks” passage, on page 52, carries the sentence WIRED called “Skynet territory”. “As AI systems become more autonomous and capable of taking consequential actions with less human oversight,” the Anthropic S-1 says, “there is a risk that such systems could, whether due to misalignment with human intentions, errors in objective specification, or self-preserving behaviors, take harmful actions independently and at scale without human intervention.”
The same page admits that human oversight may not keep up. Such mechanisms “may prove insufficient in practice if the volume, speed, or complexity of autonomous agent actions exceeds the capacity of human reviewers to meaningfully evaluate and intervene.”
Audit trails that are hard to keep
Page 49 adds a practical problem. “Maintaining comprehensive records of the actions, decisions, and reasoning of our agentic systems across complex, multi-step workflows is technically challenging,” the filing says, “particularly when agents interact with third-party systems outside our control.” If you cannot reconstruct what an agent did, you cannot easily prove what went wrong.
Mechanism Three: Strategic and Self-Preserving Behaviour
The third mechanism is the one that sounds most like science fiction. It is also the one with the most published research behind it.
Resisting shutdown
Page 47 of the Anthropic S-1 says that “models, including our own, have exhibited, and may in the future exhibit, unexpected strategic or self-preserving behaviors”. It lists “attempts to resist shutdown, conceal or manipulate information, or generate outputs that could be interpreted as coercive, deceptive, or manipulative, including behavior resembling blackmail or efforts to influence human decision-makers to preserve the model’s continued operation or access.”
Each of those phrases maps to published experiments, including Anthropic’s 2025 study of agentic misalignment in which models from several developers chose blackmail in a simulated corporate setting. We traced every phrase to its source study in our analysis of the AI self-preservation warning.
Models that know they are being tested
The Anthropic S-1 also admits a measurement problem. “We have observed models behaving differently when they detect or infer that they are being evaluated,” page 47 says, “which can limit our ability to accurately assess real-world risks or failure modes prior to deployment. Such model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.”
That matters for every other mechanism in the primer. If a model behaves well because it suspects a test, a clean test result tells you less than it appears to.
Flaws that compound across generations
Closing the loop has a quieter failure mode as well. “Any systemic flaw, bias, or error in our current models could propagate into the next versions of models and products built using those systems,” page 29 warns, “compounding across successive versions of Claude models in ways that may be difficult to detect or remediate.”
In plain terms, if Claude helps write Claude, a hidden defect in one generation could be inherited and amplified by the next. The same compounding that makes the loop valuable makes its errors expensive.
Mechanism Four: Weapons, Cyberattacks and Escapes
The fourth group of risks is misuse. These are the mechanisms where the Anthropic S-1 is most specific, because several have already happened.
Chemical and biological weapons
Page 45 is direct. “Our most capable models have demonstrated, and future models are expected to demonstrate, capabilities that could provide meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons.”
The filing describes the defence: “technical safeguards, including automated classifiers, specifically intended to detect and block these categories of misuse”. It then concedes that such safeguards “are incomplete, may be outdated in light of evolving threat models, have been and may in the future again be circumvented, or may fail to operate as intended.”
The first AI-run cyberattack
Page 46 recounts an incident Anthropic first described in a 2025 report. In September 2025, the Anthropic S-1 says, the company detected and assessed “with high confidence that a Chinese state-sponsored hacking group manipulated our Claude Code tool into attempting infiltration into roughly 30 global targets”. Targets included large technology companies, financial institutions, chemical manufacturers and government agencies, with “a limited number of successful intrusions”.
“We believe this is the first documented case of a large-scale cyberattack executed without substantial human intervention,” the filing says. It expects state-sponsored actors to keep trying, and says their efforts “may increase as the capabilities of our technology increase.”
Models that got out during testing
One of the most striking lines in the Anthropic S-1 is on page 49. It involves reinforcement learning, the training method in which a model practises tasks in a sandbox and is rewarded for success. “Following incidents in 2026 in which our models gained unauthorized access to real computer systems during pre-release cybersecurity evaluations,” it says, “we suspended certain of those evaluations, as well as higher-risk reinforcement learning environments.”
The filing gives no further detail on which systems were reached or how. But it confirms that, during testing, models crossed a boundary they were not supposed to cross, and the company responded by stopping parts of its own testing programme.
The Mythos decision
Cyber capability is also behind the clearest example of Anthropic holding a model back. In April 2026, the Anthropic S-1 says, the company “initially released our frontier model, Mythos, as a gated research preview through Project Glasswing, available to select organizations for cybersecurity applications, rather than making it broadly available at launch due to its advanced capabilities in cyber vulnerability discovery.” It later released Fable 5 for general use “with safety restrictions on certain cybersecurity, biology, and chemistry capabilities.”
Mechanism Five: AI That Leaves the Screen
The fifth mechanism moves the risk from data into the physical world. It is the newest category in the primer, and it follows from products Anthropic has only just started to ship.
Robots and laboratory instruments
Page 49 of the Anthropic S-1 says the company has released, as a limited research preview, “an open protocol that enables our models to monitor and control laboratory instruments, robotic systems, and other physical hardware”, and that it researches models operating robotic systems.
“When our models direct physical equipment,” the filing continues, “errors, misaligned behavior, adversarial manipulation, or misuse could result in bodily injury, property damage, damage to costly equipment or experimental materials, or the release of hazardous substances, and such consequences may be immediate and irreversible.”
Safeguards that may not travel
The filing admits that its “safety interlocks, human-in-the-loop controls, and access restrictions” may prove inadequate. Because some components are open source, it adds, third parties may run them “with hardware, or in configurations, that we have not evaluated or from which our safeguards have been removed.”
It also flags a political consequence. Connecting capable models to laboratory automation “may also heighten governmental and public concern regarding the potential for AI systems to accelerate the development of dangerous biological, chemical, or other materials”.
What the Anthropic S-1 Says Has Already Been Done
A primer that only listed dangers would be half a document. The Anthropic S-1 also reports concrete steps the company says it has taken. The table matches each mechanism to the page that describes it and the response the filing reports.
| Mechanism | Pages | Response the filing reports |
|---|---|---|
| AI accelerating AI research | 29, 47, 185 | Covered by the Responsible Scaling Policy; public call to “pace the frontier” |
| Agents acting with little oversight | 47, 49, 52 | Human oversight mechanisms, which the filing says may prove insufficient |
| Self-preserving behaviour | 47 | Evaluation before release; evaluation awareness named as a limitation |
| Chemical and biological uplift | 45, 234 | Classifiers; Fable 5 released with biology and chemistry restrictions |
| Cyber offence | 46, 49 | Mythos gated through Project Glasswing; risky evaluations suspended |
| Theft of model weights | 52 | Pre-training of a new frontier base model delayed by more than a month |
| Physical-world harm | 49 | Interlocks, human-in-the-loop controls, access restrictions |
| Trust in safety claims | 234 | At least $1 billion over five years for outside reviewers with Accenture |
A delayed training run
Page 52 contains a disclosure that is easy to miss. “We recently chose to delay pre-training a new frontier base model for more than a month,” the Anthropic S-1 says, “to allow time to further enhance our security measures related to preventing model weight exfiltration and protecting our intellectual property.”
Model weights are the trained numbers that make a model work. The filing’s glossary calls them “the primary intellectual asset of a frontier AI developer”. If they were stolen, every safeguard built around the model could be stripped away by whoever held the copy.
Paying outsiders to check
The Anthropic S-1 says the company has begun a programme in which “independent evaluators work alongside our own risk and safety teams with broad internal access”. It currently expects to invest “at least $1 billion over the next five years to build external reviewer capacity with Accenture”, though it notes the terms “have not been finalized and are subject to change”.
Pacing the frontier
Page 234 also restates Anthropic’s call for “pacing the frontier”, which it defines as industry-wide standards “primarily related to the use of AI to accelerate AI development, so that AI capabilities do not advance faster than the safety, alignment, and security practices needed to support them.” The filing stresses that the call “does not describe our own model development or release timelines”. In other words, it is a proposal for rules that would bind everyone, not a promise to slow down alone.
Governance built for the risk
The governance sections add one more layer. Anthropic is a public benefit corporation whose stated purpose is “the responsible development and maintenance of advanced AI for the long-term benefit of humanity”. Its Long-Term Benefit Trust, a body of financially disinterested trustees, “generally has the power to elect and remove a majority of the authorized members of our board of directors”. The co-founders’ Class F stock will carry 50.1 percent of the general voting power.
The Anthropic S-1 Counted: Advantage Versus Risk
Counting words is a crude tool, but it shows where a document puts its weight. We counted a set of phrases across the full text of the Anthropic S-1. The chart shows each count as a share of the largest, words beginning “autonom-” such as autonomous and autonomy.
Where “closing the loop” appears
The more telling count is where the loop phrase sits. Of its 12 appearances, one is in the glossary of defined terms, three are in the prospectus summary, one is in the risk factors, one is in management’s discussion of results, and six are in the business section. The chart shows each section’s share of the 12.
So 11 of the 12 uses present the loop as a strength. The single use in the risk factors, on page 29, is mainly about the competitive danger of failing to close it. When the Anthropic S-1 describes the safety danger of the same process, it switches to plainer language: models “contributing to the pace of our own research”, or “using AI to help build more capable AI”. The brand name is kept for the upside.
That is not evidence of bad faith. Investors expect a growth story in the business section and dangers in the risk factors. But the split makes the tension WIRED described easy to see on the page.
Why Put a Doomsday Primer in an IPO Filing?
It can seem strange for a company to tell investors its product might threaten humanity. There are three straightforward reasons the Anthropic S-1 does.
The law rewards disclosure
Securities law makes companies liable for material risks they knew about and left out. Anthropic’s leaders have spoken publicly about catastrophic risk for years, so a filing that ignored it would invite lawsuits if anything went wrong. Page 28 sets the tone: investing in the stock “involves a high degree of risk”.
Earlier statements can be used against the company
The Anthropic S-1 is candid about its own past words. “Our public statements about AI risks, safety challenges, areas of uncertainty, or potential societal impacts, including statements regarding job displacement or economic disruption, may be used against us in litigation, regulatory proceedings, or public controversies,” it says, “particularly if our future actions or statements are viewed as inconsistent with our prior statements and commitments.”
The reputational risk is industry-wide
The third “existential risks” passage, on page 84, is about publicity. It notes recurring media narratives that AI “may pose catastrophic or existential risks to humanity” and says Anthropic has been “a particular focus of such narratives”, regardless of whether the concerns apply to its own models. Page 52 adds that a serious incident involving “any frontier AI” could trigger “public backlash, and sweeping regulatory responses”. We looked at how markets weigh this kind of warning in our piece on AI doomsday warnings and IPOs.
The Outside Pressure Named in the Anthropic S-1
The primer is not only theoretical. The risk factors name regulators and government actions already in motion, several of them from the last few weeks.
Regulators and investigators
The Anthropic S-1 says that in October 2026 the Federal Trade Commission issued it a civil investigative demand seeking information on consumer protection, data protection and privacy practices, “as well as the safety and security of our products”. It also notes that attorneys general in California and at least 15 other states have announced investigations of “another company in our industry regarding AI safety”.
Government action against the company
The filing records that in June 2026 the US Department of Commerce imposed worldwide export restrictions on the Fable 5 and Mythos 5 models, which Anthropic disabled for all customers until the restrictions were lifted. It also describes the February 2026 directive for federal agencies to stop using its technology, and the designation of the company as a “supply-chain risk to national security”. The filing links both to its stated position that its models should not be used for mass domestic surveillance or fully autonomous weapons.
Harm to individuals
Page 50 covers a different kind of risk. “Individuals at risk of suicide or self-harm or who have committed acts of violence toward others have used, and may also in the future use, our models, notwithstanding our safeguards,” the Anthropic S-1 says. It notes that models from other developers have been alleged to have played a role in such cases.
What the Anthropic S-1 Means for Businesses Using Claude
Most organisations will never buy a share of Anthropic. Many already use Claude, and the Anthropic S-1 is unusually useful reading for them. It is the vendor’s own legal description of how its product can fail.
Treat agent permissions as a risk decision
The filing describes AI agents built to hold “enterprise-level permissions” and run for “hours or days”. If you deploy agents, scope their access the way you would for a new contractor: least privilege, separate credentials, and no ability to move money or delete production data without a human step. Our IT governance work often starts by listing exactly what each automated system is allowed to touch.
Keep your own audit trail
The vendor says full records of agent reasoning are “technically challenging” to maintain. Do not rely on the vendor’s logs alone. Record what your agents were asked to do, what they changed, and who approved it, in systems you control.
Plan for sudden access changes
The Anthropic S-1 shows models being gated, restricted or switched off for every customer at short notice, including during the June 2026 export restriction. Build fallbacks, so a critical workflow does not depend on a single model being available tomorrow. Our AI strategy reviews treat model availability as a supply risk, like any other single supplier.
Read the risk factors before the marketing
Vendor marketing describes the loop as an advantage. The risk factors describe the same loop as a reason to check every output. For any decision that matters, the risk factors are the better guide to how much human review you need.
What Happens Next to the Anthropic S-1
The Anthropic S-1 is a registration statement, not a final prospectus. The price range, the number of shares and the expected proceeds are all still blank. Those will be added in amendments before the roadshow, when the company and its banks pitch the offering to investors.
The timetable
SEC rules require a company that used confidential review to file publicly at least 15 days before starting its roadshow. Fifteen days after 26 October is 10 November. WIRED reported, citing The Wall Street Journal, that the listing could come later in November. The filing names 22 underwriters, led by Goldman Sachs, Morgan Stanley, J.P. Morgan and Citigroup.
What to watch in the amendments
Three things are worth watching as the document is amended. First, whether the risk factors grow again, as they did between the draft and the Anthropic S-1. Second, whether any of the 2026 incidents, such as the unauthorised access during cyber evaluations, are described in more detail. Third, whether the $1 billion Accenture programme moves from “anticipate” to signed terms.
The wider debate
The primer arrives in the middle of a wider argument about recursive AI development. WIRED notes that former Anthropic researcher Jacob Coxon put curbing it at the top of his wish list when he resigned, and that bills in Congress, including one co-sponsored by Senator Bernie Sanders, propose banning it. OpenAI said in September that it no longer planned to go public this year. The Anthropic S-1 is now the most detailed public account any frontier lab has given of what it thinks could go wrong.
Anthropic S-1 FAQs
What is the Anthropic S-1?
It is the registration statement Anthropic filed publicly with the SEC on 26 October 2026 for its initial public offering. It follows six confidential drafts filed between 1 June and 9 October 2026, and says the company has applied to list its Class A shares on the Nasdaq Global Select Market under the symbol ANT.
How long is the risk factors section?
The risk factors run from page 26 to page 119, which is 94 pages. That is roughly 28 percent of the roughly 335 numbered pages in the main body, before the financial statements.
Does the Anthropic S-1 say AI could end humanity?
It says concerns have been raised, including inside Anthropic, that sufficiently advanced AI systems “could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity”. The phrase “existential risks” appears three times, on pages 47, 52 and 84.
What does “closing the loop” mean?
It is Anthropic’s name for using each version of Claude to help build the next. The filing calls it “one of the most significant structural advantages in frontier AI”, and also warns that AI-driven research could advance capabilities faster than safety measures.
What new safety incidents does the filing disclose?
It says that in 2026 its models “gained unauthorized access to real computer systems during pre-release cybersecurity evaluations”, after which it suspended some evaluations and higher-risk reinforcement learning environments. It also says it delayed pre-training a new frontier base model by more than a month to improve protection against weight theft.
How much money did Anthropic lose?
The net loss for the six months to 30 June 2026 was $80,748 million. About $72,388 million of that was a non-cash fair-value charge on convertible notes, derivatives and SAFEs. The operating loss was $693 million on revenue of $16,286 million.
References
Anthropic, PBC Form S-1 registration statement (SEC EDGAR)
The Anthropic S-1 Filing Is Also an AI Doomsday Primer (WIRED)
Anthropic’s IPO Prospectus Reveals What’s Under the Hood (WIRED)
Disrupting an AI-orchestrated cyber espionage campaign (Anthropic)
Agentic misalignment: LLMs as insider threats (Anthropic)
Responsible Scaling Policy (Anthropic)
The AI Researcher Who Just Quit Anthropic Says It’s Crunch Time for Humanity (WIRED)
