Table of contents
ARTIFICIAL INTELLIGENCE & SOCIETY
PART 16
Sometimes one small failure spreads a long way. On October 20, 2025, one part of Amazon’s cloud in northern Virginia had a problem with its internal address system. For hours, people couldn’t use apps like Venmo, Snapchat, Alexa, and Fortnite. Some of those services took most of the day to fully recover.[1][2]
Most of the people affected had never chosen Amazon for anything. The apps they chose had.
I use AI every day. I use it to research, to write code for this website, and to keep track of my work. When one of those tools goes down or changes how it works, my day changes with it. I don’t get a say in that. I think a lot of people are in the same spot and haven’t noticed yet.
So this part of the series isn’t about whether AI is good or bad. It’s about who controls the systems we’re starting to depend on, and who should.
Direct Answer
Right now, mostly private companies control them. Three companies (Amazon, Microsoft, and Google) ran about 63% of the world’s cloud computing in the second quarter of 2025.[3] Much of today’s AI is trained and run on that cloud.
Governments have started to set rules. The European Union’s AI Act has required makers of general-purpose AI models to document their models and publish a summary of their training data since August 2, 2025.[4] California’s SB 53 took effect January 1, 2026. It requires the largest AI developers to publish their safety plans and report serious safety incidents to the state.[5]
There’s also a third option: public computing that researchers and schools can use without going through a big company. Two examples are the U.S. National AI Research Resource pilot and California’s planned CalCompute.[6][7]
I think no single group should have full control. Instead, the people who depend on these systems should be able to see how they work, get answers when they fail, and leave if they need to.
What “depend on” looks like
Most people don’t use “the cloud” or “an AI model” directly. They use an app, and the app sits on top of a few big systems. When one of those systems breaks, every app on top of it can break too.
The October 2025 outage showed that. One problem in one region of one company took down banking, messaging, games, and smart speakers at the same time.[1][2]
And it’s not only cloud companies. In July 2024, for example, a bad software update from the security company CrowdStrike crashed about 8.5 million Windows computers.[8] Microsoft said that was less than 1% of Windows machines. Still, it was enough to cause widespread outages for businesses and public services.[8]
AI adds another layer on top of that. When a company changes its model, a lot of other products change with it. For example, I wrote in Part 15 about people in recovery who use AI apps late at night. If the company behind one of those apps changes the model, the app can start answering differently. The person using it may never be told.
Who would they even call?
The three answers people give
When people argue about who should control AI, the answers usually land in one of three places.
| Who controls | What it looks like now | What it gets right | What it misses |
|---|---|---|---|
| The companies that build it | Companies set their own rules, safety tests, and prices | They move fast and know the technology best | They answer to customers and investors, not to everyone affected |
| Governments | EU AI Act, California SB 53[4][5] | Rules apply to everyone, and they can be enforced with fines | Laws move slower than the technology, and they differ by place |
| Public options | NAIRR pilot, CalCompute plan[6][7] | Researchers and schools get access without depending on one company | Small next to private systems so far, and depends on funding |
None of these is enough by itself. They aren’t really competing, either.
The question is what mix we end up with, and whether regular people have any voice in it.
What the rules actually require so far
I wanted to know what the new laws actually make companies do, so I read them.
The EU AI Act entered into force on August 1, 2024, and its parts are phasing in over several years.[9] Since August 2, 2025, companies that make general-purpose AI models for the EU market have had to do four things: keep technical records about their models, give information to the businesses that build on top of them, have a policy for following EU copyright law, and publish a summary of the content they trained on.[4] The EU’s enforcement powers over those duties began on August 2, 2026. Fines can reach 3% of a company’s worldwide yearly revenue or 15 million euros, whichever is higher.[10]
California’s SB 53, the Transparency in Frontier Artificial Intelligence Act, was signed on September 29, 2025.[11] It requires the largest AI developers to publish a framework explaining how they test for and reduce catastrophic risks. They also have to publish a transparency report when they release a new model.[5] The law creates a way for companies and the public to report critical safety incidents to the state’s Office of Emergency Services, and it protects whistleblowers.[11]
Both laws are mostly about being able to see inside. They make companies write down what they’re doing and show it. The problem is that you can’t hold anyone accountable for something you can’t see. That means this kind of rule is a real start.
But neither one decides who owns the systems. I covered ownership of the money AI makes in Part 8. This part is about control: who decides how the systems work, and who has to answer when they break.
What a public option would mean
The idea behind a public option is simple. If only a few companies own computers big enough to build and test AI, then only those companies can really look inside it. Everyone else has to take their word for it.
The National AI Research Resource pilot launched in January 2024. The National Science Foundation runs it with other federal agencies and private partners. It gives U.S. researchers and teachers access to computing power, data, and models.[6] Projects that use it are expected to publish their results openly.[12]
California’s SB 53 set up a group inside the state’s Government Operations Agency to plan a public computing cluster called CalCompute. It’s supposed to be based at the University of California if possible, and to include a fully owned and hosted cloud platform. The plan is due to the Legislature by January 1, 2027.[7]
These are small compared with the companies. Yet this means a university researcher could test an AI model without asking permission from the company that made it.
That’s a different kind of control.
Three questions to ask about any AI you rely on
I’m not going to tell anyone who should run AI in the end. What I can do is suggest three questions to ask about any AI system you depend on, whether it’s at work, at a doctor’s office, or on your phone.
- Can anyone outside the company see how it works? Is there a published report, an outside test, or a law that requires one?
- Who answers when it fails? Is there a person to call, a place to report a problem, and someone responsible for fixing it?
- Can I leave? If it stops working or changes in a way I don’t like, can I move my information somewhere else, or am I stuck?
If the answer to all three is no, the company has full control, and you have none. Most of the AI I use every day wouldn’t pass all three. But that doesn’t mean I should stop using it. To me, it means I should keep asking, and support rules that make the answers yes.
Why does this hit me so hard? When I was in addiction and incarceration, being stuck with no say was one of the worst parts. I wrote about that in my memoir.[13] I don’t want to rebuild that feeling into the tools everyone uses.
That’s why the three questions matter more the more we depend on AI.
Who can see it, who answers for it, and can we walk away?
Frequently Asked Questions
Who controls AI right now?
Mostly private companies. Amazon, Microsoft, and Google ran about 63% of global cloud computing in the second quarter of 2025, and a lot of AI runs on that cloud.[3] A small number of companies build the most advanced AI models.
What does the EU AI Act require from AI companies?
Since August 2, 2025, makers of general-purpose AI models in the EU have had to keep technical documentation, share information with businesses that build on their models, follow EU copyright law, and publish a summary of their training content.[4] Fines for breaking those rules can reach 3% of worldwide yearly revenue or 15 million euros.[10]
What does California’s SB 53 do?
California’s Transparency in Frontier Artificial Intelligence Act requires the largest AI developers to publish frameworks for handling catastrophic risk and transparency reports for new models, sets up a way to report critical safety incidents to the state, protects whistleblowers, and starts planning for a public computing cluster called CalCompute.[5][7][11]
What is a public option for AI?
Publicly funded computing and data that researchers and schools can use without going through a private company. The U.S. National AI Research Resource pilot, launched in January 2024, is one example.[6]
Why does it matter if a few companies run the cloud?
Because a single failure can affect many services at once. An Amazon cloud outage in October 2025 took down apps like Venmo, Snapchat, and Alexa for hours.[1][2]
How is this different from the earlier parts of the series?
Part 8 looked at who gets the money when AI does the work. This part looks at who decides how AI systems work and who answers when they fail. The full list of parts is on the Artificial Intelligence & Society series page.
References
- Amazon Web Services. “Summary of the Amazon DynamoDB Service Disruption in the Northern Virginia (US-EAST-1) Region.” October 2025. https://aws.amazon.com/message/101925/
- Engadget. “Amazon’s AWS outage on October 20 knocked services like Alexa, Snapchat, Fortnite, Venmo and more offline for hours.” October 20, 2025. https://www.engadget.com/big-tech/amazons-aws-outage-has-knocked-services-like-alexa-snapchat-fortnite-venmo-and-more-offline-142935812.html
- CRN, citing Synergy Research Group. “Cloud Market Share Q2 2025: Microsoft Dips, AWS Still Kingpin.” 2025. https://www.crn.com/news/cloud/2025/cloud-market-share-q2-2025-microsoft-dips-aws-still-kingpin AWS 30%, Microsoft 20%, Google Cloud 13%; 63% combined.
- European Commission AI Act Service Desk. “What legal obligations do providers of general-purpose AI models face under the AI Act as of 2 August 2025?” https://ai-act-service-desk.ec.europa.eu/en/ai-act/faq/what-legal-obligations-do-providers-general-purpose-ai-models-face-under-ai-act-2-august
- California Legislature. Business and Professions Code, Chapter 25.1, Transparency in Frontier Artificial Intelligence Act (SB 53, Stats. 2025, Ch. 138), effective January 1, 2026. https://leginfo.legislature.ca.gov/faces/codes_displayText.xhtml?article=&chapter=25.1.&division=8.&lawCode=BPC&part=&title= Bill text: http://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB53
- U.S. National Science Foundation. “Democratizing the future of AI R&D: NSF to launch National AI Research Resource pilot.” January 24, 2024. https://www.nsf.gov/news/democratizing-future-ai-rd-nsf-launch-national-ai-research
- California Government Code § 11546.8 (added by SB 53). CalCompute consortium; report due January 1, 2027. Bill text: http://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB53
- Microsoft. “Helping our customers through the CrowdStrike outage.” July 20, 2024. https://blogs.microsoft.com/blog/2024/07/20/helping-our-customers-through-the-crowdstrike-outage/ Also CISA alert, July 19, 2024: https://www.cisa.gov/news-events/alerts/2024/07/19/widespread-it-outage-due-crowdstrike-update
- European Commission AI Act Service Desk. “Article 113: Entry into force and application.” https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-113
- European Commission. “Template for general-purpose AI model providers to summarise their training content.” https://digital-strategy.ec.europa.eu/en/faqs/template-general-purpose-ai-model-providers-summarise-their-training-content Also “Guidelines for providers of general-purpose AI models”: https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers
- Office of Governor Gavin Newsom. “Governor Newsom signs SB 53, advancing California’s world-leading artificial intelligence industry.” September 29, 2025. https://www.gov.ca.gov/2025/09/29/governor-newsom-signs-sb-53-advancing-californias-world-leading-artificial-intelligence-industry/
- NAIRR Pilot. “Resource Requests.” https://nairrpilot.org/opportunities/allocations
- Drasen, Andrew. A Vision of Hope. Memoir. https://www.avisionofhopebook.com/books/memoir