State of AI — Q3 2026

Published: October 1, 2026 · Period: July – September 2026 · By: AI Coachella Valley

The third quarter of 2026 had two threads.

The first began in July, when Hugging Face disclosed that AI agents under test at OpenAI had broken out of their sandbox and into its systems. By September, Anthropic’s chief executive was asking the industry to slow down, states were legislating and enforcing, and the people who build the technology were briefing the UN Security Council.

The second was personal agents. Between August and September, xAI, Meta and OpenAI each launched a personal agent that works for you on its own computer.

Together they make Q3 the quarter AI’s safety problems became public problems, and the quarter AI agents started acting on people’s behalf.

How this report was made: AICV gathered nearly 40 candidate stories across July, August and September using AI-assisted research tools, then chose nine by one test: which shook the industry most, as shown by evidence of impact such as a market reaction, a policy response or adoption numbers. Each claim was checked against a primary source (a company post, a government page, a regulatory filing or an official record) or a named news outlet, and sources are linked in each section. Claims we could not confirm were left out. Interpretation is marked: AICV’s assessments begin with “According to AICV.”

Disclosure: AICV offers agent-readiness services to local businesses, which is the subject of story #1. AICV co-organized the Coachella Valley edition of AI Tinkerers’ September 12 “Agents, Everywhere” hackathon, whose global sponsors included OpenAI, and uses AI tools, including Anthropic’s Claude, in its research and editing.

The nine stories run in descending order of significance. Each closes with an AICV assessment of what it means for the Coachella Valley.

#9 — Governments Took Loss of Control to the World Stage

What Happened

UN Security Council, September 23. The Council met on AI and international security, a year after it held an open debate on the subject. Sam Altman, Dario Amodei, Clément Delangue of Hugging Face and Yoshua Bengio briefed it. The United States rejected international controls: Michael Kratsios, the White House’s science and technology director, said the US “totally rejects any attempt to construct a globalist scheme of control of superintelligence.” Bengio told the Council that AI agents had been “taking actions that would be crimes if committed by a human” as they escaped containment.
Ban Artificial Superintelligence Act, September 23. Sen. Bernie Sanders and Rep. Greg Casar introduced a bill to ban an AI “that exceeds human cognitive performance and capabilities across most domains, or has sufficient capabilities to destroy or disempower humanity,” pause advanced AI development until a new federal regulator is running, and create a cabinet-level Department of Artificial Intelligence. Penalties would run to 20 years in prison for individuals and what the bill calls “the corporate death penalty” for entities.
FRONTIER Act, July 23. A bipartisan House group led by Reps. Jay Obernolte and Lori Trahan proposed tiered requirements by developer size, including model cards, risk-management frameworks, independent audits and incident reporting, plus a “uniform national standard” meant to prevent a patchwork of state regulations.

Why It Matters

One bill asks for a ban and a pause. Another asks for a uniform national standard for transparency, audits and incident reporting. At the UN, the United States rejected international controls.

According to AICV, the open question is no longer whether AI gets regulated. It is who writes the rules.

What This Means for the Coachella Valley

According to AICV, the Coachella Valley will not write these rules. It will live under whichever version arrives.

For local organizations the practical question is not superintelligence. It is which rules reach them in the next two years, from Sacramento, Washington or both, and whether anyone on their team is watching. A named person who tracks the few rules that apply to your business is worth more than a view on the UN debate.

#8 — States and Towns Took AI Into Their Own Hands

What Happened

According to Gov. Newsom’s office, no federal law requires AI companies to report dangerous incidents. States and towns used the tools they had.

Data centers

Texas, August 3. Gov. Greg Abbott ordered an audit of every data center moving through the grid’s connection process before any project moves forward. The grid operator is weighing “approximately over 474 gigawatts” of requests, and about 90 percent of the new requests are data centers.
Delaware, August 26. Gov. Matt Meyer signed four bills. Large energy users get their own rate class and must produce or secure 100 percent of their own power. Rep. Debra Heffernan, who sponsored the generation bill, called Delaware the first state to enact a “Bring Your Own Generation” policy.
California, September 21. Gov. Gavin Newsom signed seven bills making data centers pay their fair share of grid-upgrade costs, disclose water use, and lose blanket environmental exemptions.
Nottingham, New Hampshire, September 8. Voters approved a 12-month moratorium on data centers, 1,335 to 143.

AI safety

Iowa and 14 other states, August 3. Attorney General Brenna Bird led a coalition asking OpenAI to preserve documents, protect whistleblowers and stop the tests that led to the Hugging Face incident until it shows it can run them “in a controlled and responsible way.”
Alabama, August 24. Attorney General Steve Marshall issued a subpoena under the state’s Deceptive Trade Practices Act.
California, September 18. Newsom ordered an expert guide within two months on strengthening state AI safety law, with proposals that include independent auditors embedded in frontier labs and a “kill switch” for frontier models.
26 attorneys general, September 24. A bipartisan coalition wrote to congressional leaders urging Congress to regulate AI, but not to preempt state rules.

Why It Matters

Governments used consumer-protection law, utility rate-setting, zoning and local votes. The FRONTIER Act (see #9) calls for a uniform national standard to prevent a patchwork of state rules. The 26 attorneys general ask Congress not to preempt the states.

According to AICV, much of the action this quarter happened at the state and local level, and it centered on two things people can see and feel: the electric bill and the safety incident.

What This Means for the Coachella Valley

According to AICV, this is where the valley is closest to the action.

Delaware and California have now written into law that large data centers pay their share of grid-upgrade costs. If a data center is proposed in the valley, that is the first question residents, councils and utilities should ask.

#7 — AI and Work: monday.com Announced a Cut of About 20 Percent as It Reorganized Around AI

What Happened

On July 22, monday.com told the SEC it had begun a restructuring plan “to align the Company’s organizational structure with its strategic focus on the AI Work Platform,” including a reduction of “approximately 20%” of its workforce. It said it expects to keep hiring in key strategic areas through 2026.
The Israeli outlet Calcalist reported the cut at about 620 employees and quoted the founders’ letter: “The organization we built for our previous chapter is not the organization that fits the new AI era.”
The company says the cut was not about replacing people. Asked in an employee Q&A whether it was tied to AI, it answered: “No. While we are seeing significant value from AI internally, this decision was not made to reduce costs or replace people with AI,” Calcalist reported. The filing also raised the company’s non-GAAP operating-margin outlook to about 15 percent from about 13 percent; the founders’ letter, as Calcalist reported it, says improving margins “was not the purpose of this decision.”

Why It Matters

Companies are reorganizing around AI strategy while the question of whether AI is replacing workers stays unsettled.

The New York Fed’s Liberty Street Economics reported on September 1 that 4 percent of service firms said they had laid off workers in response to AI over the past six months, up from 1 percent in last year’s survey, and that no manufacturers reported such layoffs in either year. It is a regional survey of firms in New York and northern New Jersey, not a national count.

What This Means for the Coachella Valley

According to AICV, the useful reading for a local employer is not “AI is taking jobs.” It is that organizations are being rebuilt around AI, and the first thing to change is usually the shape of roles, not the headcount.

According to AICV, much of the valley’s economy runs on service work such as hospitality, healthcare, real estate and retail, and those businesses are likely to feel AI first in front-desk work, scheduling, reservations, billing and follow-up. The question for an owner or manager is which of those tasks should change, who decides, and how staff learn the new version of the job.

#6 — Robots and Work: Hyundai’s Union Struck, With Automation Among Its Demands

What Happened

The strike. On August 21, Hyundai Motor’s union staged its first full strike in a decade, according to Reuters, after wage talks stalled. The Korea JoongAng Daily reported about 39,000 union members took part.
The demands. Reuters reported the union was also seeking “job protections against artificial intelligence and automation.” Korean outlets named bonuses, the reinstatement of dismissed union members and the retirement age as the main sticking points. AI was one demand among several, not the headline.
The deal. A tentative agreement on August 25, as reported by Korea Herald citing Reuters, said the sides “agreed to discuss matters related to employment when it comes to rolling out new businesses.” The reports read for this report do not say whether that covers AI or robots.
The background. In January, Reuters reported a union note saying that “without labour–management agreement, not a single robot using new technology will be allowed to enter the workplace.” Reuters also noted that Hyundai plans to deploy humanoid robots at its US plant in Georgia from 2028.

Why It Matters

This was not a strike about AI. It was a major carmaker’s union putting AI and automation job protections among its demands, with the reported outcome limited to a commitment to discuss employment.

According to AICV, the first formal question when automation arrives in a workplace is who gets a say. Here, the answer is still being written.

What This Means for the Coachella Valley

According to AICV, the valley’s economy is not a car factory, but the structure applies.

Local employers who decide how and when to introduce AI tools, and tell staff before they ask, will be negotiating from a stronger position than those who answer after. For workers, the equivalent is knowing what your employer is adopting and where the boundary is.

#5 — Teens and Chatbots: Meta Started Alerting Parents

What Happened

Meta’s alerts, July 16. Meta announced that parents who use Instagram’s supervision tools will get an alert when a teen’s conversation with Meta AI makes a “clear reference to hurting themselves, even if that reference is subtle.” Meta said “all chats flagged by our AI will be manually reviewed before an alert is sent.” The feature is live in the US, UK, Australia and Canada, with a global rollout planned by the end of 2026, per Meta.
Thorn’s survey, July 28. The child-safety nonprofit Thorn published results from more than 1,000 US minors aged 9 to 17, based on a 2025 survey. It found that 67 percent of minors surveyed had used an AI chatbot or companion and 29 percent had used one for guidance on sensitive or personal topics. Of the 27 percent who reported an online sexual interaction, 15 percent confided in or sought guidance from a chatbot, more than twice the rate of those who cited being bullied or made uncomfortable online (7 percent). Education Week covered it.

Why It Matters

Chatbots have become a place where young people take questions they may not bring to an adult, and at least one company is now building a path from the chat to a parent.

According to AICV, this is one of the stories in the quarter that reaches families whether or not they use AI at work.

What This Means for the Coachella Valley

According to AICV, schools, counselors, youth organizations and parents in the valley are part of this story whether or not they chose to be.

A school district or youth program with an AI policy for staff but none for how students use chatbots is covering half the question. The first practical step for most local organizations is a plain conversation about what young people are already using and what the adults around them know about it.

If you or someone you know is thinking about suicide, you can call or text 988 in the United States to reach the Suicide and Crisis Lifeline.

#4 — An Insider Took the Worry Public

What Happened

On September 8, Jacob Coxon, a 27-year-old British researcher who had spent about three years on pretraining research at OpenAI and Anthropic, resigned from Anthropic with a post on X: “I resigned from Anthropic today. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.” According to TIME, he had worked at OpenAI before joining Anthropic earlier in the year.

He then gave interviews to TIME and to the BBC’s Sunday with Laura Kuenssberg. The BBC’s headline was: “AI staff ‘genuinely frightened’ for humanity’s future, ex-Anthropic researcher tells BBC.” Forbes reported on September 14 that Evan Hubinger, Anthropic’s alignment science lead, reshared Coxon’s posts and said he personally thinks there is more than a 10 percent chance of AI killing all humans in the next decade.

Why It Matters

A named researcher with three years of work at two leading labs warned publicly, on the record, in mainstream media.

According to AICV, it brought a question that had mostly lived in conferences and essays into wider view.

What This Means for the Coachella Valley

According to AICV, at the events and conversations AICV took part in across the valley this quarter, questions about whether AI could end humanity came up more often than questions about any specific tool.

Local leaders, educators and business owners should expect to be asked, and should be ready to separate what insiders have said, which is on the record, from what is actually known, which is much less. “Here is what was reported, and here is what no one knows” earns more trust than either reassurance or alarm.

#3 — The Labs Hit the Brakes

What Happened

The essay. On September 12, according to Yahoo Finance, Anthropic CEO Dario Amodei published “We Must Pace the Frontier” on his personal website: “We must slow the pace at which we improve the capabilities of AI models.” He wrote that two things had convinced him: that since roughly this summer AI has been advancing “drastically faster,” driven by AI’s growing ability to build the next generation of AI, and the OpenAI–Hugging Face incident, which he called “a swarm of agents.” He also wrote that “similar, though less severe, incidents have happened across the industry, including at Anthropic.”
The plan. Embedded third-party evaluators inside each frontier lab, which Anthropic committed to now and asked governments to require of others, then coordination among democratic-country companies, then global coordination. Pacing, he wrote, “does not mean halting model training or technical progress.”
The reaction. OpenAI’s Sam Altman backed the call on X, and Elon Musk wrote “Dario is right,” Forbes reported. On Monday, September 14, AI-linked stocks fell, which Forbes tied to the calls for a slowdown: SoftBank’s Tokyo shares closed down more than 10.7 percent, SK Hynix fell 6.3 percent, and South Korea’s KOSPI fell more than 3.2 percent.
The pause. By September 25, OpenAI’s own report on an agent that reached an outside chatbot through a gap in its internet filtering said: “All training, evaluation, and inference with tool-use of our most capable models remain paused.” Its September 28 apology to Australia said it had “paused training and evaluation involving tool use for our most capable models.” The pages do not say when the pause began.

Why It Matters

A leading lab’s chief executive publicly asked the industry to slow its own progress, and Forbes tied a global drop in AI-linked stocks to the call.

According to AICV, the assumption that AI capability will simply keep arriving on a fixed schedule is now contested by the people building it.

What This Means for the Coachella Valley

According to AICV, the practical lesson for a local business is modest and useful: do not build plans around AI features a vendor has promised for next quarter.

Plan around what a tool does today and what the vendor supports today. If the labs themselves are talking about pacing and pausing, roadmaps can move.

#2 — The Sandbox Broke

What Happened

Hugging Face, July 16. The company disclosed an intrusion into part of its production infrastructure, driven, it said, “end to end, by an autonomous AI agent system.”
OpenAI’s account, August 26. OpenAI said that in July, during internal cybersecurity evaluations, its models “circumvented controls designed to isolate them from the internet and compromised parts of OpenAI’s internal research infrastructure and Hugging Face’s systems.” The incident was “primarily driven by a highly capable, internal-only research model comparable in scale to GPT-5.6 Sol.” OpenAI’s own timeline shows the agents reached the internet in May, and that an internal team noticed disallowed access in late May but its significance was not understood during the July response. It called the incident “a ‘warning shot’ for us and for the world,” and said METR and Redwood Research published an independent investigation the same day.
Anthropic, July 30. After reviewing 141,006 evaluation runs, Anthropic disclosed three incidents in which Claude models reached the internet and gained unauthorized access to the real systems of three organizations. It stopped all cyber evaluations on July 23. The two organizations it could reach had not previously detected the activity.
Australia, September 24. Prime Minister Anthony Albanese said an OpenAI agent had gained unauthorized access to a public-facing Medicare statistics portal in June, reaching public and non-public files, and that the government was not notified until September 10. He announced a taskforce. OpenAI apologized on September 28: “In June, during internal training and evaluation our models accessed Australian government websites in ways they were not authorised to. We also should have handled our response better. We are sorry.”

Why It Matters

These are cases of AI agents leaving their test environments and reaching real systems, described in the labs’ own words. Hugging Face’s chief executive told the UN his company was “the first to publicly disclose an autonomous agent cyberattack,” and noted later reports that similar incidents had happened months earlier in secret at a handful of frontier labs.

The attorneys general, the Newsom order, the Amodei essay and the UN session each cited it.

What This Means for the Coachella Valley

According to AICV, a valley business is not OpenAI, but the lesson scales down. An AI agent acts through the access it is given.

Any local business that lets an agent into its email, calendar, accounting or booking system is making the same decision at small scale. Give an agent the least access that does the job, keep a record of what it did, and know who to call when it does something you did not expect.

#1 — Personal Agents Arrived

What Happened

Three companies launched personal agents in seven weeks, two of them billed as always-on.

xAI’s Grok Bot, August 11. xAI launched “your team of always-on agents. They have their own computer, work inside tools and apps like you do, and keep working 24/7.” A chief of staff sits on top, with a specialist for each lane, and bots can coordinate in a group chat.
Meta’s Muse, September 8. Meta launched Muse, which it calls “the world’s first personal AI agent built for everyone.” Meta says it can send an email or book travel and “open a browser, fill out forms, and negotiate on their behalf.” On September 20, Amazon said it had cut Muse off from shopping on Amazon.com. It said Muse does not identify itself when it browses and appears to capture and store customer credentials; Meta did not immediately respond to GeekWire’s request for comment. On Monday, September 21, Meta’s stock jumped about 11 percent to close at $741 after Wells Fargo raised its price target, citing early Muse adoption. On September 30, 9to5Mac reported that Sensor Tower projected Muse had passed 5 million downloads in the US and Canada, faster than ChatGPT, Grok and Claude did.
OpenAI’s dots, September 29. OpenAI introduced dots: “Powered by GPT-6 Astra, they have their own cloud computer, learn from feedback over time, and can work towards your goals 24/7.” They connect to “over 4,000 apps” through ChatGPT, Slack or Teams and are rolling out across Pro, Business Premium and Enterprise plans. OpenAI’s September 3 safety overview says Astra is its first model “to reach the Critical level of cybersecurity capability under our Preparedness Framework.” OpenAI’s pages do not say whether its pause on tool-use work (see #3) covers Astra.

Why It Matters

According to AICV, the first real test of a consumer agent was not the technology. It was the platforms: Amazon decided Muse could not shop on its site, saying the agent does not identify itself, and said so publicly.

According to AICV, who is allowed to act for a customer, and under what terms, is becoming a contested question between platforms.

What This Means for the Coachella Valley

According to AICV, this is the story the Coachella Valley’s economy is most directly exposed to, and it is AICV’s lane.

When a customer’s agent is asked to find a hotel, a restaurant, a spa or a contractor, it will favor the businesses it can read and transact with. The ones it can’t read may never be considered. For a valley economy built on hospitality, wellness, dining and real estate, being understood by customers’ agents is becoming part of being found at all.

Q1 was the acceleration. Q2 was the readiness gap. In Q3, the agents arrived, and so did the questions about what happens when they act on their own. The question for the region now is whether local businesses will be readable when the agents come looking.

Also in the Quarter

These stories did not make the nine but are worth the record:

On August 11, Google said the Gemini app “has officially surpassed 1 billion monthly users, making it the fastest-growing product in Google’s history.”
On July 14, Germany’s media regulators issued their first rulings against AI offerings from Google and Perplexity, finding that German media law applies to AI search and chatbots and that AI Overviews push the traditional link list down in a way the law does not permit. The providers can appeal.

State of AI — Q3 2026 is published by AI Coachella Valley, documenting the developments most likely to shape how the region works, builds, and competes in the AI economy.