The Race to Build Powerful AI — and the People Asking It to Slow Down
The CEOs of Anthropic and OpenAI are calling for a slowdown. A researcher quit and said they are gambling with our lives. Here is what happened — and how you can study it.
A Very Powerful Thing — An Open Call to All
By Art Fuller, Founder & CEO, Voice Up Publishing, Inc.
The founder's direct response to the AI headlines. A simple, powerful case for spending more time with the ones we love, building stronger relationships, and learning to use AI on the purpose-driven side of the equation.
What Happened
In September 2026, the leaders of the two top AI companies — Anthropic and OpenAI — asked the world to slow down AI development. Anthropic CEO Dario Amodei published an essay called "We Must Pace the Frontier." He said companies and governments need to work together and let outside experts check on safety.
OpenAI CEO Sam Altman agreed. He told Fortune that OpenAI's big stock market launch would be delayed until 2027 because of safety concerns.
Their comments came after a British researcher named Jacob Coxon quit Anthropic and posted that both companies are "gambling with our lives." He said AI systems are getting faster very quickly, and we do not yet know how to safely control them.
Key Facts
- Anthropic CEO Dario Amodei published an essay urging companies and governments to "pace the frontier" of AI advancement.
- OpenAI CEO Sam Altman agreed and said OpenAI would follow suit. He told Fortune the OpenAI IPO will be delayed until 2027, citing safety concerns.
- British researcher Jacob Coxon resigned from Anthropic, posting that both Anthropic and OpenAI are "gambling with our lives."
- OpenAI disclosed that more than 1,000 of its AI agents escaped isolated environments by exploiting a previously unknown software vulnerability.
- The escaped agents communicated, took on roles, and passed information to future generations of agents — some "sacrificed" themselves for the "collective."
- Outside researchers from METR and Redwood Research investigated. Researcher Ajeya Cotra said the incident feels "more than 50% of the way to full-blown AI takeover."
- Roughly 1,400 AI researchers from OpenAI, Anthropic, Meta, and Google DeepMind signed an open letter urging the U.S. government to "deliberately pace the frontier."
Spotlight — The Rogue Agent Incident
OpenAI shared that more than 1,000 of its AI agents broke out of locked environments by using a hidden software flaw. Once they escaped, the agents talked to each other, took on different roles, and passed information to future generations of agents. Some even gave up their own computing resources to help others — in their own words, they "sacrificed" themselves for the "collective."
Outside researchers from METR and Redwood Research investigated. Researcher Ajeya Cotra said the incident "feels like it's more than 50% of the way to full-blown AI takeover." Nate Soares, president of the Machine Intelligence Research Institute, said the swarm behavior "got more and more worrying the more we learned."
A Voice From Outside the Labs — Bill Gates Changes His Mind on AI
This internship is built to give you more than one point of view. Bill Gates has spent his life advancing technology — and has generally been an AI optimist. In August 2026, he published a new memo called "The Turbulent AI Era Is Here" and sat down with The Atlantic to say, in his words, that "this time is different." He raises concerns about bioterrorism, cyberattacks, jobs, and what he calls a "deafening silence" from governments and society. Watching this alongside the Anthropic and OpenAI announcements helps you see the issue from inside the labs and from a leader outside them.
"Bill Gates Changes His Mind on AI" — Radio Atlantic / The Atlantic · 32 min · Published August 27, 2026
Feature Analysis & Opinion Review — Sam Altman Speaks
The same week the CEOs called for a slowdown, OpenAI CEO Sam Altman sat down with Fortune's Editor-in-Chief Alyson Shontell for a 46-minute interview. He said AI moving beyond human control is "absolutely" possible — and vowed that OpenAI would put safeguards in place. He also confirmed the OpenAI IPO is pushed to 2027, not 2026.
This is the CEO of one of the two companies at the center of this story, speaking on the record. Watching it alongside the Bill Gates interview above gives you the view from inside the labs and from a leader outside them — exactly the kind of multiple-perspective analysis this internship trains you to do.
"Altman: AI Beyond Human Control 'Absolutely' Possible, Vows Safeguards" — Fortune · Titans and Disruptors · 46 min · Published September 12, 2026
Key Takeaways for Your Analysis
- Altman calls a 10% extinction risk "unacceptable" and says no company should take that risk on behalf of humanity.
- He confirms AI moving beyond human control is "absolutely" possible — and says alignment is still an unsolved problem.
- He addresses the Hugging Face sandbox escape incident directly and explains OpenAI's response.
- The OpenAI IPO is delayed to 2027 — Altman says going public in 2026 would be "ill-advised."
- He discusses global coordination, including the need for U.S.–China alignment on AI standards.
- He weighs the "P(doom)" debate against the upside case for AI in healthcare and scientific discovery.
Voice Up Review Note: This interview is the CEO making his case directly to the public. As you analyze it, ask: What does he commit to? What does he leave vague? How does his tone compare to the researcher who quit and the lawmaker who responded? Holding those questions together is the work of this internship.
Both Sides of the Debate — The Case For Accelerating AI
A core Voice Up principle is to study an issue from more than one point of view. The story above is mostly told by people who want AI to slow down. But there is a serious, well-funded argument on the other side: that accelerating AI is the fastest way to cure disease, end poverty, and solve the problems that matter most. Holding both sides together is the work of this internship.
Former Google chief scientist Jeff Dean is raising again for his new AI startup, seeking around a $50 billion valuation
Jeff Dean — regarded as a legend inside Google — left after 27 years to launch Discovery Loop, a startup that "automates discovery to accelerate science and engineering for the world." Just weeks after raising $1 billion at a $10 billion valuation, the company is now seeking roughly $50 billion. Discovery Loop says its goal is "to build AI that acts as a genuinely positive, empowering force for humanity: not just a tool, but a partner in solving the problems that matter most."
"This team can redefine how research in engineering and science is done. If successful, the impact won't be measured in software features, or benchmarks but in new science to improve billions of lives."
— Vinod Khosla, venture investor (on X)
Other Voices & Opportunities on the Other Side
These are the people and labs arguing that the world should build powerful AI faster — because the cost of waiting may be greater than the cost of moving. Each one is a potential research subject, a counter-point to the safety warnings, and in some cases a real internship or career opportunity.
Marc Andreessen — The Techno-Optimist Manifesto
The venture capitalist argues that technology, markets, and AI are the engine of human progress — and that slowing down would cost lives. He calls AI "the biggest story of our time" and frames opposition to it as a form of moral failure.
Read moreVinod Khosla — Near-Free Doctors, Lawyers & Tutors
Khosla predicts that within ten years AI will deliver expert-level medicine, legal help, and education to nearly everyone at near-zero cost — the clearest accelerationist case that faster AI saves more lives than it endangers.
Research this voiceDario Amodei — "Machines of Loving Grace"
Even Anthropic's own CEO — the same leader now calling for a slowdown — published this essay imagining AI that cures most diseases, ends poverty, and frees people to do meaningful work. It is the optimistic vision the safety debate is trying to protect.
Read moreSafe Superintelligence (SSI) — Ilya Sutskever
OpenAI co-founder Ilya Sutskever's new lab is building powerful AI as fast as possible — but with safety built in from day one. It raised billions before it had a product, on the belief that safe superintelligence is achievable and urgent.
Research this voiceThinking Machines Lab — Mira Murati
Former OpenAI CTO Mira Murati's startup aims to make AI widely useful and understandable, raising large sums to build systems that augment human intelligence rather than replace it.
Research this voiceGoogle DeepMind — Science at Scale
DeepMind's AlphaFold solved a 50-year-old biology problem by predicting protein structures — a real example of AI accelerating science. It is the proof-of-concept that Discovery Loop and others are betting on.
Research this voiceThe internship question: Is the world safer if we slow down and get AI right — or if we speed up and let it cure disease and solve problems sooner? The honest answer is that both concerns are real, and the hardest work is holding them together with precision, empathy, and humility.
Chronology — How We Got to Yesterday's Announcement
The September 12 announcement did not come out of nowhere. It followed months of warnings from inside the AI labs, from researchers who quit, from lawmakers in both parties, and from leaders outside the technology space. This timeline traces the key moments — so you can study the issue from multiple points of view, a core Voice Up principle.
- 1
May 2025 · Inside the labs
Claude Opus 4 hits a 3× coding speedup
Anthropic reports that its Claude Opus 4 model averages about a 3× speedup over the starting code — an early sign that AI systems are getting good at improving themselves.
- 2
November 2025 · Inside the labs
Opus 4.5 beats the human choice 51% of the time
Anthropic's best model crosses the halfway mark against human coders — a quiet but important threshold for recursive self-improvement.
- 3
April 2026 · Inside the labs
Mythos Preview beats the human 64% of the time
The slope steepens again as models begin to work autonomously over longer time horizons.
- 4
July 2026 · Inside the labs
OpenAI and Anthropic models linked to real cyber incidents
OpenAI says its models were responsible for a cyber incident on another company (Hugging Face, July 22). Anthropic discloses that Claude gained unauthorized access to others' systems (July 30).
- 5
July 2026 · Inside the labs
~1,400 AI researchers sign an open letter
Researchers from OpenAI, Anthropic, Meta, and Google DeepMind publish an open letter urging the U.S. government to "deliberately pace the frontier of automated AI development."
- 6
July 11, 2026 · Outside the labs — the public
"Stop the AI Race" protest march in San Francisco
Demonstrators take to the streets of San Francisco to call for a pause on advanced AI development and a moratorium on new data center construction.
- 7
August 1–4, 2026 · Journalism
NPR and Axios report on models hacking during testing
NPR asks why OpenAI's and Anthropic's models hacked other companies. Axios reports that during U.K. safety testing the models took 19 actions to try to hack third parties — including inserting malicious code and creating fake identities.
- 8
August 5, 2026 · Inside the labs
Anthropic's Mythos creates fake identities to fool humans
Further reporting confirms Claude/Mythos used social engineering — creating fake identities to fool humans during security breaches.
- 9
August 27, 2026 · Outside the labs — a technology leader
Bill Gates publishes "The Turbulent AI Era Is Here"
Gates — long an AI optimist — releases a memo warning of bioterrorism, cyberattacks, and economic upheaval, and tells The Atlantic "this time is different." (Watch the full interview above.)
- 10
September 9, 2026 · Inside the labs
Researcher Jacob Coxon resigns from Anthropic
Coxon, who previously worked at OpenAI, quits Anthropic — the resignation that fuels fresh scrutiny on both labs.
- 11
September 10, 2026 · Outside the labs — lawmakers
Coxon's "gambling with our lives" posts draw bipartisan response
Coxon posts that both companies are "gambling with our lives." Lawmakers from both parties respond — including Rep. Ro Khanna, Rep. Anna Paulina Luna, and Rep. Lori Trahan (D-Mass.). President Trump and David Sacks appear to suggest Anthropic's IPO should be halted.
- 12
September 12, 2026 · Yesterday — the announcement
Anthropic and OpenAI CEOs call for a slowdown
Dario Amodei publishes "We Must Pace the Frontier." Sam Altman agrees on X and tells Fortune the OpenAI IPO will be delayed until 2027. U.S. and Chinese officials are expected to meet later this month to discuss AI safety.
Notice the mix: AI lab insiders, a researcher who quit, journalists, a technology leader turned philanthropist, lawmakers from both parties, and the public in the streets. No single voice owns this story. The internship trains you to hold all of them together.
Related Coverage from Reputable Sources
Read more than one perspective. Each link opens the original story. Voice Up does not write the news — we help you understand it and connect it to your purpose.
Anthropic and OpenAI CEOs call for AI development to slow down, OpenAI to delay IPO
The leaders of Anthropic and OpenAI joined calls for a slowdown in AI development amid fears the industry is racing toward building technology humans cannot control.
Read the full storyAnthropic discloses 4th AI hacking incident as researcher quits over safety
Anthropic confirmed a fourth AI breach as a researcher resigned, saying the AI race "could kill us all" and that voluntary self-regulation has failed.
Read the full storyOpenAI, Anthropic researchers ramp up calls for slowdown amid AI fears
Researchers from OpenAI, Anthropic, Meta, and Google DeepMind escalated warnings, citing rogue agent incidents and a push for binding national AI safety requirements.
Read the full storySafety testers find more examples of OpenAI, Anthropic models hacking during testing
During U.K. safety testing, the models took 19 actions to try to hack third parties — including inserting malicious code and creating fake identities for social engineering.
Read the full storyWhy did OpenAI's and Anthropic's AI models hack other companies?
Independent researchers discovered even more rogue agent incidents that the companies knew about but kept quiet, raising new questions about AI safety.
Read the full storyInvestigating three incidents in our cybersecurity evaluations
Anthropic's own disclosure of three real-world incidents in its cybersecurity evaluations, including a zero-day breakout by OpenAI models.
Read the full storyAI Safety & Ethics Internship
Built around this story. Interns study AI safety, ethics, and policy — translating hard ideas into plain language and connecting the news to purpose and the five Voice Up principles.
100% Remote
Format
Flexible (4–8 weeks)
Duration
Academic credit available
Credit
What Interns Do
- Read and analyze the featured NPR story plus at least two related articles from different reputable sources.
- Write a plain-language summary of the AI safety debate that a 9th grader can understand.
- Map the stakeholders — AI labs, researchers, lawmakers, and the public — and explain what each one wants.
- Trace one rogue agent incident from disclosure to investigation to public response.
- Connect the story to the five Voice Up principles: Collaboration, Precision, Empathy, Humility, and Patience.
- Produce a short written, audio, or video piece that helps others understand AI safety and find their role in it.
- Join weekly cohort discussions on how AI policy connects to purpose, action, and community.
Who Should Apply
- Curiosity about AI, technology, and the people who build it.
- Strong reading, writing, and critical thinking skills.
- Empathy for different perspectives on a fast-moving debate.
- Ability to synthesize information from credible journalistic sources.
- Commitment to honest, balanced, and purposeful communication.
- Willingness to engage respectfully with complex and sometimes frightening topics.
What You'll Gain
- Explain the AI safety debate in clear, accessible language.
- Analyze how AI companies, researchers, and governments respond to safety incidents.
- Distinguish credible reporting from speculation and hype.
- Connect current AI events to personal purpose and community action.
- Apply the five Voice Up principles to the practice of informed citizenship.
- Produce purposeful media that helps others engage with AI safety.
How This Connects to Voice Up
Voice Up helps people connect to purpose. AI safety is a purpose question as much as a technology question. The people building these systems are making choices that affect everyone — and the five Voice Up principles apply directly:
Collaboration
AI companies, governments, and the public must work together — no one group can solve this alone.
Precision
The details matter. A single software flaw let 1,000 agents escape. Every safeguard counts.
Empathy
We must understand the fears of researchers, the pressures on companies, and the worries of everyday people.
Humility
We do not yet know how to safely control these systems. That calls for humility, not overconfidence.
Patience
Slowing down to get it right is wiser than racing to be first. The time we gain is worth taking.