> daily_signal(2026_05_27)

Sam Altman walked back the AI jobs apocalypse in Sydney the same week the FBI started filing data-center protesters under terrorism.

PickBits Daily Signal · Wednesday, May 27, 2026

By Mark Pickering · 7 min read · May 27, 2026

// tl;dr

Sam Altman told a Commonwealth Bank of Australia audience in Sydney he was wrong about the AI jobs apocalypse he himself had predicted. "I'm delighted to be wrong about this." It is the second walk-back in eight days from a frontier-lab CEO; Nvidia's Jensen Huang called the same framing "just too lazy" on Tuesday. Anthropic's Chris Olah is the dissent, telling the Vatican that displacement at scale is still "a real possibility" and a "moral imperative."
Wired published a thousand-plus pages of unreleased DHS, FBI, and fusion-center reports grouping climate activists, anti-Israel protesters, and data-center critics under a new category called "anti-tech violent extremism." A Northern Virginia fusion center lists photography of an AI data center as suspicious activity. The label is now part of the federal threat lexicon.
Bloomberg confirmed Beijing has expanded passport-surrender and exit-ban requirements from DeepSeek staff (where it started in March) to Alibaba, Manus, Moonshot AI, and StepFun. AI researchers at named private firms now need state approval to travel abroad. Manus co-founder Xiao Hong and chief scientist Ji Yichao are among the restricted individuals. China is treating AI talent the way it treats nuclear scientists.
Samsung's 48,000-member chip-workers union ratified the first AI-revenue profit-share deal in industry history with 73.7 percent in favor and 95.5 percent turnout, vote closed 10 a.m. Korea time today. The bonus pool is 10.5 percent of semiconductor-division operating profit, paid only to Device Solutions staff. The non-chip divisions got $4,000 each and have a shareholder lawsuit going. The 10.5 percent number is the floor the next union to negotiate will start at.
The UK Academy of Medical Royal Colleges told the British government to treat children's access to AI chatbots the way the country treats smoking and seatbelts. The Academy speaks for 22 royal colleges and made the recommendation in its formal submission to the consultation that closed yesterday. Doctors Manitoba's parallel survey: 90.4 percent of doctors back restricting AI chatbots and social media for minors. Legislation is expected before year-end.

Yesterday Nvidia's Jensen Huang called the AI-blamed-for-layoffs framing "lazy" and "irresponsible." Today Sam Altman in Sydney said he was wrong about the jobs apocalypse he himself had predicted. The two CEOs whose companies sit on either end of the AI cost curve, the chips and the model, walked away from the prediction inside eight days of each other. At the same time the federal government quietly filed a thousand-plus pages of reports treating data-center protesters as a counterterrorism concern, Beijing made its AI researchers surrender their passports, Samsung's chip workers ratified the first AI-revenue profit-share deal in industry history, and the British medical establishment told the UK government to put AI chatbots for kids in the same category as cigarettes. The buildout did not stop today. The narrative around who is harmed by it, who benefits from it, and who is allowed to oppose it shifted on five fronts at once.

Today Sam Altman in Sydney said the jobs apocalypse he predicted did not happen, the FBI started writing reports about people who photograph data centers, Beijing took AI researchers' passports, Samsung chip workers ratified a 10.5 percent profit-share bonus, and British doctors told their government to treat AI chatbots for kids like cigarettes.

1. Sam Altman in Sydney admitted he was wrong about the AI jobs apocalypse. Jensen Huang said the same thing on Tuesday. Anthropic's Chris Olah told the Vatican the opposite.

Continuing 5.26 #6 (Jensen Huang's "lazy" rebuke) and 5.25 #1 (the Vatican AI doctrine launch with Chris Olah on stage).

At a Commonwealth Bank of Australia conference in Sydney on Tuesday, OpenAI CEO Sam Altman told Matt Comyn on stage: "I'm delighted to be wrong about this," referring to his own previous prediction that AI would compress white-collar job turnover into a much shorter window. "I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened." He added that OpenAI was "roughly right" on the technological predictions made at ChatGPT's 2022 launch but "pretty wrong" on social and economic impact. The Sydney comments come eight days after Altman's own AI Pulse "Where We Are Now" essay had still flagged white-collar displacement as a near-term risk. Fortune's lead read the Sydney remarks as IPO positioning; Time's read them as a course correction that lines up with what the labor data has actually done since 2023.

On Tuesday Nvidia CEO Jensen Huang told CNA that CEOs blaming layoffs on AI are "just too lazy to figure out how to do it right," called the framing "irresponsible," and named Meta, Wix, and Standard Chartered directly. Today's Altman walk-back is the second from a frontier-lab CEO in eight days, and the two of them sit on opposite ends of the AI cost curve, the chip side and the model side. The dissent is from Anthropic co-founder Chris Olah, who told an audience at the Vatican's Magnifica Humanitas encyclical launch yesterday that AI displacement of human labor "at very large scale" remains "a real possibility" and supporting affected workers will become "a moral imperative of historical proportions." Two CEOs walked back. One co-founder doubled down. The labor narrative inside the frontier labs is now publicly split for the first time.

Euronews coverage of Sam Altman in Sydney saying no AI jobs apocalypse so far — euronews.com · May 26, 2026

Why this matters: If your boss used AI as the reason for a layoff this year, or floated it as the reason for one coming, the CEOs of OpenAI and Nvidia just publicly said the framing is wrong, eight days apart. That is the chip and the model, the two companies most invested in selling you the opposite story. The cover Altman wrote with the original prediction is gone; the cover Huang demolished on television is gone. The dissent is from Anthropic, not from Nvidia or OpenAI. Action this week: Save the Altman Sydney quote and the Huang CNA quote (Euronews and Tekedia have the cleanest write-ups). The next time your company runs an all-hands where AI is named as a workforce factor, the question to ask the speaker is the one Huang asked: AI has been useful for six months; which named six-month-old AI tool replaced which named worker on which named team? If your function is hiring right now and an interviewer cites AI as the reason for restructuring, the two quotes are the politest way to ask for specifics; an honest answer names the tool and the workflow, a dishonest one pivots to the road map.

euronews.com: No AI 'jobs apocalypse' so far, says OpenAI's Sam Altman

2. Wired published a thousand pages of FBI and DHS reports labeling data-center protesters "anti-tech violent extremism." Photography of an AI data center is on the suspicious-activity list.

Wired reported Tuesday on more than 1,000 pages of unpublished documents from the FBI, the Department of Homeland Security, and regional fusion centers, obtained by the magazine, that describe a new federal threat category called "anti-tech violent extremism." The category groups climate activists, anti-Israel protesters, and people opposing the construction of AI data centers under a single label. A Northern Virginia Regional Intelligence Center bulletin reviewed by Wired lists "expressed/implied threat," "observation/surveillance," "photography," "testing/probing of security," and "attempted intrusion" as suspicious-activity indicators in the data-center context. Photographing an AI data center is now, on paper, an activity a fusion center will write a report about.

The documents cite a string of incidents to justify the category: the April 6 shooting at the home of Indianapolis city-county councilmember Ron Gibson after he voted in favor of a contested data center (13 bullet holes, a note reading "No Data Centers"), the Molotov firebombing attempt at Sam Altman's residence, and the November 2025 arson by an insurrectionary anarchist cell at the Equinix AI campus in Meudon, France. Civil-liberties groups quoted by Wired and Gizmodo flagged that the category is broad enough to fold in residents of Saline Township, Morgan County, and the dozens of other counties currently in active opposition to a data-center build. Marjorie Taylor Greene publicly criticized the framing on Tuesday, calling the surveillance of AI critics a civil-liberties failure. The criticism came from both directions in the same news cycle, which is the unusual part of this story.

Android Authority coverage of US law enforcement now monitoring for anti-tech extremism in connection with AI data center opposition — androidauthority.com · May 26, 2026

Why this matters: If your county is one of the dozens currently fighting an AI data-center project, the federal threat lexicon now has a category that can be applied to people on your side of that fight. The fusion-center bulletin lists ordinary investigative behavior, photographing a site, watching it, asking questions, in the same column as actual violence. That does not mean every protester is being tracked. It means the category exists, and once a category exists the bar to applying it tends to drop. The Indianapolis shooting was real; so was the Meudon arson. The folding of those into a frame that also covers showing up to a town hall with a camera is the part civil-liberties groups are flagging. Action this week: If you are involved in local data-center opposition, read the Wired piece in full and the Android Authority and Gizmodo summaries for context. Document your activity on paper, keep meeting minutes, file written comments through the official channels (town council, county planning, your state PUC). The paper trail is what separates a community organizer from a target of a fusion-center bulletin. If you work in IT or compliance, the same documents flag observation of corporate sites as a watch indicator; have your physical-security and legal teams reread their tipline-handling policy with this lexicon in mind, because the calls are going to start coming.

androidauthority.com: Law enforcement is now monitoring for 'anti-tech extremism'

PickBits Daily Signal is the working brief I file every morning at 6 a.m. ET. If a friend forwarded this to you and you want it in your inbox, you can subscribe at pickbitsai.substack.com. It is free.

3. Beijing made AI researchers at Alibaba, DeepSeek, Manus, Moonshot, and StepFun surrender their passports. Bloomberg confirmed the expansion today.

Bloomberg reported today that the Chinese government has expanded its overseas-travel restrictions for top AI talent from DeepSeek, where the rule started in March, to staff at Alibaba, Manus, Moonshot AI, and StepFun. Researchers, executives, startup founders, and engineers at the named firms have been asked to surrender their passports to their employers and to apply for state approval before traveling abroad. The formal justification is that the work could give access to information classified as a state or commercial secret. Two co-founders of Manus, Xiao Hong and chief scientist Ji Yichao, are among the individuals confirmed by Bloomberg as currently restricted. The restrictions are individual and do not require formal charges.

The escalation is on top of an April 2026 directive from the National Development and Reform Commission requiring any AI firm to obtain government approval before accepting US investment, which already pulled in Moonshot AI and StepFun. The combined policy treats elite AI staff the way Beijing already treats nuclear scientists and senior executives at state-owned firms. The trigger Bloomberg cites is March's DeepSeek R1 release, which briefly upended assumptions about how much compute Chinese frontier labs needed to reach Silicon Valley benchmarks. Stanford's 2026 AI Index put the gap between the best US and Chinese models at 2.7 percent, down from 17.5 to 31.6 percentage points in mid-2023. The policy direction is one Beijing has been moving toward for a year; the named-firm expansion is the part that landed today.

The Next Web coverage of China expanding AI travel curbs from DeepSeek to other private firms — thenextweb.com · May 26, 2026

Why this matters: If you work for a company that licenses, distributes, or invests in any Chinese AI model or service, the named-employee constraints just became personal. Researchers behind those models are now being held at a national-security threshold for cross-border movement. Conference attendance, collaboration trips, academic exchanges, and US partnership talks at the named firms all have a new approval gate that did not exist last week. If you use a Chinese model on your phone or in a product, none of those features go away today; the constraint is on the people behind them rather than on the outputs you see. The strategic mirror is the US export-control regime, and the two regimes now constrain each other's people directly. Action this week: If your company uses a Chinese AI service (Alibaba Cloud, Tongyi Qianwen, DeepSeek's API) or is talking to one, pull up the current vendor list and write down which firms now have employee-mobility constraints attached to them. The Bloomberg piece (paywalled), The Next Web and Cryptobriefing summaries, and Stanford's 2026 AI Index are the four pieces to read in sequence. If you are a researcher attending an upcoming conference (NeurIPS, ICLR, ACL), check whether the speakers from named firms are confirmed in person versus remote; the pattern over the next 12 months is the signal worth tracking.

thenextweb.com: China extends AI travel curbs from DeepSeek to other private firms

4. Samsung's 48,000 chip workers ratified the first AI-revenue profit-share deal in industry history. 73.7 percent yes, 95.5 percent turnout. [Returning]

Continuing 5.23 #1 (the Day-1 74 percent vote and parallel shareholder lawsuit and mobile-division court challenge). The ratification result landed today as that arc's resolution.

The National Samsung Electronics Union (NSEU) ratification vote on the 10.5 percent AI-revenue profit-share deal closed at 10 a.m. Korea time on Wednesday, May 27. The result: 73.7 percent in favor of 46,142 votes cast, on a turnout of 95.5 percent of 65,593 eligible members. The deal directs a special bonus pool funded by 10.5 percent of operating profit from the Device Solutions semiconductor division to chip-workers only, with up to 600 million won (about $400,000) per individual at the top end. Average wages across the union rise 6.2 percent and a housing-loan program is introduced. The strike that had been scheduled for late May is officially off.

The internal tensions did not disappear with the result. Non-chip Device eXperience (DX) employees, who work on phones, displays, and consumer products, received bonuses of about $4,000 and have an active production slowdown going in the chip-packaging line. A shareholder lawsuit filed earlier this week seeks to block the underlying 26.6 trillion won (about $26 billion) payout on fiduciary grounds. The lawsuit was cleared from the immediate injunction track but the merits case is pending. Tom's Hardware's read on the result is that the precedent now exists in a major Asian semiconductor company: union members can negotiate a percentage of AI-driven operating profit as a wage-floor mechanism, indexed to the segment that captures the AI revenue. The 10.5 percent figure is the number every other named union in the next 12 months will reach for.

Korea Herald coverage of Samsung strike-averting wage deal passing union vote — koreaherald.com · May 27, 2026

Why this matters: If you work for a company whose revenue is going up because of AI and your paycheck is not, the Samsung union just established that a percentage of the AI-driven operating profit can be written into the wage agreement as a separate pool. The number is 10.5 percent of segment operating profit, paid to the staff doing the work that generates it. That number now exists as a public benchmark. Whether your country has unions, works councils, or just an HR department, the question "what share of the AI-driven profit goes back to the people building it" has a first answer attached to it. Action this week: Save this so the next time you negotiate compensation at a company that is publicly reporting AI-driven revenue growth, you have a real-world reference point to anchor against. If you are in HR, finance, or executive comp, the Korea Herald and Xinhua write-ups have the deal structure; the 10.5 percent pool is calculated on segment operating profit (not total revenue, not gross), which is the part that will get negotiated everywhere else. If you sit on a works council in Germany, the Netherlands, France, or the Nordics, the structure is portable to the European framework with one substitution (the European version typically negotiates as a percentage of an EBITDA line rather than US-style operating profit).

koreaherald.com: Samsung's strike-averting wage deal passes union vote

5. UK doctors told their government to treat children's access to AI chatbots the way the country treats smoking.

The Academy of Medical Royal Colleges, which represents 22 royal medical colleges across the United Kingdom, submitted its formal recommendation to the government's "Growing Up in the Online World" consultation, which closed yesterday. The submission frames the harms to children from social media and AI chatbots as comparable in scale to smoking. The Academy's specific recommendations included age-based restrictions on AI chatbots for under-16s, film-style age ratings for social media platforms, bans on personalized services for under-13s, night-time curfews, and a disabling of auto-play and infinite scroll. More than half of 132 doctors the Academy surveyed reported seeing at least one case of health harm related to tech and devices every week; over a third reported multiple cases a week. The recommendations sit alongside Doctors Manitoba's parallel survey of Canadian physicians, in which 90.4 percent support restricting social media and AI chatbots for minors.

The political context is the part that travels. Australia banned social media for under-16s in December 2025; France, the EU member states, and Brazil are pursuing similar restrictions; the UK government has said it expects to legislate before year-end. With the AMRC submission, the first major Western medical-establishment body has gone on record placing AI chatbot access for kids in the same regulatory framework as cigarette age restrictions, seat-belt mandates, and food-safety inspections. The framing is what changes here. Once an issue has a "doctors-vs-tobacco" template attached to it, the political momentum tends to follow. The US GUARD Act, which advanced through the Senate Judiciary committee on a 22-0 vote on April 30, is the closest US analogue; the Hawley-Cruz reconciliation and the House companion bill are still pending.

Medscape coverage of UK doctors warning social media poses smoking-level health risk to children — medscape.com · May 26, 2026

Why this matters: If you have a kid under 16 in the UK, an age-based AI chatbot restriction is now in front of the British government with the backing of the country's medical-establishment umbrella body. If you have a kid in the US or Canada, the political momentum behind the same framing just got named, with a percentage attached (90.4 percent of Canadian doctors surveyed). Whether the legislation lands in three months or fifteen, the public-health template the AMRC just used is the one that ended indoor smoking; it is unusual for AI chatbots to get attached to that template this fast. Action this week: Open Family Link (Android), Screen Time (iOS), or the equivalent on your kid's device and check what AI chatbot apps are installed. If you do not know which ones include chatbots (TikTok, Snap, Instagram, and Character.AI all do, in different ways), the Common Sense Media app reviews are the cleanest reference. Set time limits where they are not set, and turn off in-app purchases on any app with a chatbot feature for under-13s. If you work in product or trust-and-safety at a consumer platform with AI features, the AMRC consultation response (filed with Department of Health and Social Care on May 26) is the document to read; the next regulatory wave in the UK will start from those recommendations.

medscape.com: Doctors Warn Social Media Poses Smoking-Level Health Risk

» What to watch this week

Q2 earnings calls that quietly drop AI from the layoff explanation. Standard Chartered already walked back Bill Winters's "lower-value human capital" phrasing. After the Altman and Huang walk-backs, the next round of major-company earnings calls (June starts the season) will tell us which CEOs are about to retire the framing and which double down.
FOIA filings on the Northern Virginia Regional Intelligence Center bulletin. The ACLU, EFF, and Brennan Center traditionally file under the federal and state FOIA regimes within days of a Wired disclosure of this size. Watch their public dockets; the litigation that follows determines whether the "anti-tech extremism" category persists with this scope or gets narrowed.
Hugging Face hosting policy on abliterated open-weight variants. Last week's Heretic disclosure put the platform on the spot. A policy change would be a precedent for what the next platform does when a similar tool ships. (Continuation of 5.26 #2.)
The first US union to publicly cite the Samsung 10.5 percent number. The UAW, CWA, and SAG-AFTRA all have active or upcoming negotiations with AI-revenue-exposed employers. The first time the Samsung figure shows up in a US proposal is the moment the precedent crosses jurisdictions.
UK government legislative timeline on the AMRC recommendations. The Department of Health and Social Care indicated it intends to legislate before year-end. Watch for a White Paper or a Statement to the House on the post-consultation findings; the AMRC submission is the document most likely to shape the first draft.
Brunswick, Maine town council data-center moratorium hearing on June 1. The first Maine municipal vote on its own pause after the statewide moratorium veto. The outcome sets the template for the rest of the state.

Tomorrow's signal lands here.