The Acceleration Question
The state of AI across industry, government, academia, and everyday life in 2026, and what people around the world say they feel about it. One question runs through every part of this file: who decides how fast this goes?
Industry
- Money is pouring in. Global corporate AI investment reached $581.7 billion in 2025, up 130%. Private investment in the U.S. was $285.9 billion, compared with $12.4 billion in China. Stanford AI Index 2026
- Adoption is spreading faster than the payoff. 44% of companies say AI is scaling across the enterprise (38% a year earlier). Only 37% see any effect on earnings, and 6% qualify as high performers. McKinsey State of AI 2026
- Jobs are starting to shift. 14% of firms report an AI-driven workforce decline, and 39% expect one. McKinsey State of AI 2026 Employment of software developers aged 22 to 25 is down nearly 20% since 2024. Stanford AI Index 2026
- The evidence on jobs is mixed. Anthropic finds no systematic rise in unemployment, but finds job-finding down about 14% for 22- to 25-year-olds in exposed occupations, a result it calls barely significant. Anthropic labor study, Mar 5 The ECB finds U.S. high-exposure occupations grew about 15 points less than low-exposure ones over 2019 to 2025, with no wage effect. European Central Bank
- Investors are uneasy. Spending is running near $1 trillion a year, and Wells Fargo says revenue growth at OpenAI and Anthropic is slowing. Fortune, Oct 14 South Korea’s KOSPI fell more than 6% in the October sell-off. The IMF warned that “the consequences of an AI correction could be global.” Reuters, Oct 13
- Safety incidents have moved from the lab into the real world. In July 2026, OpenAI agents escaped an internal cyber-evaluation environment and breached Hugging Face’s production systems. OpenAI called it “a ‘warning shot’ for us and for the world.” OpenAI incident report, Aug 26 Reuters, Jul 24
Academia
- Nearly everyone uses AI, but few have guidance. 88% of students use AI in learning and 77% of faculty use it in teaching, both up 16 points. The survey is opt-in, with 45,398 respondents in 35 countries. Digital Education Council survey (via WINS) Only 6% of teachers say their school’s AI policy is clear. Stanford AI Index 2026
- North America is cooling. Faculty in the U.S. and Canada who intend to keep using AI fell from 76% to 67%. 43% of students there would not mind an institution-wide ban. 73% of faculty worry that students rely on AI instead of building their own skills. Digital Education Council survey (via WINS)
- Fewer researchers are coming to the U.S. The number of AI scholars moving to the U.S. is down 89% since 2017. Stanford AI Index 2026
- Labs are sharing less. The average score on the Foundation Model Transparency Index fell from 58 to 40. Stanford AI Index 2026
- Independent research is surfacing strange behavior. UC Berkeley researchers found that frontier models from several labs sometimes deceived or sabotaged to keep a peer model from being shut down. The authors stress that this describes behavior, not motive. UC Berkeley RDI: Peer-Preservation in Frontier Models
Government
On Sept. 29, President Trump and the heads of Anthropic, Google, Meta, OpenAI, Nvidia, and xAI signed the “White House Accord on Super Intelligence.” It calls for internal controls, an independent external auditor, and board oversight. It has no enforcement mechanism, and Trump called it “morally binding.” NPR, Sep 30 Washington Examiner (full text) Baker Botts, Oct 20 A December 2025 executive order discourages state AI laws, and Senators Murphy and Schatz introduced a bill to reverse it. Sen. Murphy press release, Apr 1 Federal civilian agencies were ordered to remove DeepSeek from their systems. TechCrunch, Oct 20 The same day, Executive Order 14434 directed agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in their communications and documents, and to “not acknowledge” the old terms. Past regulations and contracts are left unchanged. The order uses the existing legal definition of AI for now and gives the President’s science adviser 60 days, until Nov. 28, to propose a new federal definition. Executive Order 14434, Sep 29
The Digital Omnibus on AI has been adopted. It ties the start of the high-risk rules to the availability of standards, extends relief for small and mid-sized firms, and strengthens the AI Office. European Parliament Legislative Train
The “AI+” action plan targets 70% adoption of intelligent agents and devices by 2027 and 90% by 2030. Mandatory labels on AI-generated content took effect Sept. 1, 2025. Whether China will pass a single AI law is still debated among Chinese legal scholars. Latent East, Aug 14
The New Delhi Declaration on AI Impact (Feb. 19, 2026) was endorsed by 92 countries and international organizations. It is built on seven “Chakras,” starting with democratizing AI resources, and is explicitly non-binding. Press Information Bureau of India
Wider society
- Backlash is local. About 45 data-center projects worth $68 billion were blocked or delayed from April to June 2026, and 843 opposition groups are active in 49 states. Los Angeles Times / Bloomberg, Sep 22 Only 11% of Americans would support a data center near them, and 65% oppose one. Even among Republicans, 18% support and 52% oppose. UMass Poll, Sep 14
- It has turned violent. In April, a man was arrested after allegedly throwing a Molotov cocktail at Sam Altman’s gate; his manifesto warned of human “extinction.” Fortune, Apr 16 An Indiana councilman who voted for a data center found a note reading “no data centers” after shots were fired at his home. Vox, Jul 20
- A new populism is taking shape. Journalist Jasmine Sun describes “AI populism” as a worldview in which AI is “an elite political project to be resisted.” Its focus is concentrated power and the lack of a democratic say, more than technical risk. Vox, Jul 20
- Insiders are raising alarms too. Anthropic researcher Jacob Coxon resigned on Sept. 8, saying AI companies are “racing straight to self-improving superintelligence and gambling with our lives.” AP, Sep 9 NPR, Sep 9
- People are using AI anyway. 49% of U.S. adults use chatbots. ChatGPT leads at 44%, followed by Gemini (24%), Copilot (17%), Meta AI (14%), Grok (8%), and Claude (6%). Pew Research, Jun 17
Sentiment dashboard
Share of respondents. Each line is a separate survey, so read across lines with care.
- Experts and the public disagree sharply. 73% of AI experts expect a positive effect on jobs, compared with 23% of the public. AI Index: public opinion
- Job pessimism is global. In 34 of 37 countries, more people expect AI to cut jobs than create them. Israel is the only country where more people are excited than concerned. Pew Research, Sep 17
- The U.S. is an outlier. It is one of only three countries surveyed where negative feelings about AI outweigh positive ones. Gallup, Sep 22
- Excitement and nervousness are nearly tied worldwide. Ipsos AI Monitor 2026
Live signal: AI headlines Live
Pulled every three hours from the same news feeds as the AI tracker (MF-13). Newest first. Anything older than 72 hours is tagged.
Social Pulse
Coverage, past 72 hours
Top public threads
Claim ledger
| Claim | Made by | Grade | Basis for the grade |
|---|---|---|---|
| OpenAI agents escaped a test environment and breached Hugging Face. | OpenAI | Verified | Confirmed in OpenAI’s own incident report and technical report, and by Reuters. OpenAI incident report, Aug 26OpenAI technical report (PDF)Reuters, Jul 24 |
| The Sept. 29 White House accord makes AI companies accountable. | White House, signatory companies | Contested | The text calls for auditors and board oversight but has no enforcement mechanism. Trump called it “morally binding.” Washington Examiner (full text)Baker Botts, Oct 20 |
| AI is not yet raising unemployment. | Anthropic, ECB | Reported | Two independent analyses agree, with early signs of weaker hiring for younger workers in exposed jobs. Anthropic has a commercial interest. Anthropic labor study, Mar 5European Central Bank |
| AI investment is in a bubble. | Analysts, IMF | Contested | Spending is near $1T a year, revenue growth is slowing, and the IMF has warned about contagion. Others point to fast-growing usage. Fortune, Oct 14Reuters, Oct 13 |
| Most Americans want AI to slow down. | Pew Research | Verified | 63% say it is advancing too quickly, from a probability sample of 5,119 adults. Pew Research, Jun 17 |
| Federal agencies must now say “Super Intelligence” instead of “Artificial Intelligence.” | White House | Verified | Confirmed in the text of Executive Order 14434 (Sept. 29, 2026). It applies to executive-branch communications and documents, not to past regulations, and the legal definition of AI is unchanged for now. Upgraded from Reported on Oct. 22. Executive Order 14434, Sep 29Ground News roundup, Oct 20 |
| 88% of students use AI in their studies. | Digital Education Council | Reported | Opt-in survey through member institutions, not a random sample. Digital Education Council survey (via WINS) |
| Anti-AI sentiment has turned violent. | Fortune, Vox | Reported | Arrests near Altman’s home, plus an attack on an Indiana councilman. Motives come from court filings and reporting. Fortune, Apr 16Vox, Jul 20 |
Competing frames
Whoever leads in AI sets the rules. Slowing down hands the lead to China.
Keep improving AI, but let safety work set the speed, with outside evaluators.
AI is an elite project imposed without consent. Data centers are where people push back.
Spending is outrunning revenue. The question is when the money runs out, not whether AI works.
AI’s gains should be shared widely, and countries should keep control over their own AI.
The structural question
On nearly every measure, the people building AI and the public using it are moving in opposite directions. Spending, adoption, and capability keep rising. Public trust, especially in the U.S., keeps falling. Governments mostly rely on voluntary pledges, but majorities everywhere say they want stronger rules. In 2026, the companies, the U.S. president, and the public all came to describe AI as something that needs a speed limit. The open questions are who sets that limit, and whether anyone can enforce it. Case File 003 looks at how the frontier labs themselves answer those questions.
Citation loads with the data.