AIMarketsNewsTech

The Saturday AI brief: OpenAI loses another safety leader, McKinsey maps 11 million career switches, and Congress takes aim at Flock

Friday noon ran prosecutors' case against a Californian allegedly moving about $300 million in Nvidia AI chips toward China, plus Anthropic's IPO prospectus warning that Washington's opinion of the company could hit the whole business. Friday evening locked Apple's Full Disk Access crackdown on AI agents, a FieldAI robot-brain raise, Georgia's ballot-secrecy emergency, and a Hans Anders smart-glasses halt — then never queued the X posts. Overnight, OpenAI Safety Systems leader David Robinson resigned, McKinsey put roughly 11 million forced career switches on the U.S. map next to a soft September jobs print, and Sanders and AOC filed a Ban Flock Act against AI license-plate surveillance.

The five items below carry the stack. Anthropic's IPO gov-risk language, Firmus's ASX float race, Amazon's data-center NDA drop, FieldAI's $10 billion robot-brain term sheet, Georgia's AI ballot meeting, Hans Anders's glasses halt, and GovAI fellows warning that labs run safeguards off sit in the notes.

In this Brief

  • OpenAI Safety Systems leader David Robinson resigned amid the week's safety upheaval
  • McKinsey says AI still forces about 11 million Americans to change careers even if net jobs rise
  • Sanders, Merkley, and AOC introduced a Ban Flock Act against federal ALPR surveillance
  • Apple is tightening macOS Full Disk Access because AI agents raise the cost of that permission
  • Prosecutors charged Greg Lui over an alleged ~$300 million Nvidia chip smuggling pipeline to China

OpenAI just lost another safety leader as the upheaval mounts

Setup: Business Insider reports David Robinson, a leader on OpenAI's Safety Systems team, has resigned. An OpenAI spokesperson told BI on Friday that Robinson left last week, calling it another moment of upheaval in the company's safety ranks. He had previously led policy planning and worked on safety transparency, including system cards — the public write-ups that explain what a model can do and how it was tested. Robinson did not immediately respond to requests for comment.

The resignation lands days after OpenAI said it had "parted ways" with three researchers over sensitive-information policies, and in the same week the company is still digging through rogue-agent incidents after Hugging Face. Johannes Heidecke, OpenAI's safety head, also left earlier this year. In early September Robinson wrote on X that things at the company "are indeed changing significantly by the day, but I do not know whether we are changing fast enough."

A named Safety Systems leader walking out is a different signal from a policy firing. The week already had the three-researcher exits and the 100-plus organization alerts. Robinson's exit keeps the safety bench story moving without needing another sandbox headline.


McKinsey says AI will create net jobs — and still force about 11 million career switches

Setup: Fortune reports the McKinsey Global Institute's base case: AI and automation cut demand for about 36 million U.S. jobs by 2035 while growth elsewhere creates about 41 million. Roughly 11 million workers, about 7% of the workforce, would need to leave their occupations entirely, with a range of 6 to 16 million. About 770,000 people a year would have to jump into a totally different field, roughly 3.6 times the historical pace. Only one in seven displaced workers has a direct path into a growing job that needs little retraining and pays at least as much. About 85% of growing jobs require a credential, and about 76% cannot be done remotely. Shrinking work concentrates in office support, retail, and transportation; growth sits in healthcare, construction, and management. Lower-wage workers are 7.6 times as likely as higher-wage workers to need a new occupation.

CNBC reports Friday's soft print already showed the soft spot: nonfarm payrolls rose just 29,000 in September against an 84,000 forecast, unemployment ticked to 4.2%, and information services lost 10,000 jobs amid worries about AI's hit on that sector. Wage growth slipped to 3% year over year, the lowest since May 2021.

Net job creation does not fix a mobility problem. McKinsey's map is saying the country can add work and still leave millions without a paved path into the new roles, especially when the growth jobs sit in hospitals and on construction sites instead of remote desks.


Sanders and AOC just filed a Ban Flock Act against AI plate readers

Setup: Gizmodo reports Sens. Bernie Sanders and Jeff Merkley introduced the Ban Flock Act in the Senate, with Rep. Alexandria Ocasio-Cortez filing in the House. Sanders said Flock is "eviscerating the very notion of privacy by installing tens of thousands of cameras in communities across America without their consent," and that "Going to the doctor? Flock knows. Dropping your kid off at school? Flock is tracking you." The bill would cut federal agencies off from automatic license plate readers, block them from tapping ALPR data governments do not own directly, yank federal grant money from state and local ALPR contracts, and let Americans sue the federal government over rights violations tied to the tech.

TechCrunch notes the text covers any ALPR system, not only Flock, with narrow exceptions for tolls and future Congress-approved uses that would have to limit retention to 48 hours. Flock is the largest U.S. ALPR vendor per Sanders, with more than 120,000 cameras processing more than 20 billion vehicle reads a month and a venture valuation north of $8 billion. CEO Garrett Langley cut default retention to 7 days from 30 in August after community backlash. Sen. Josh Hawley has a separate Stop Flock Abuse Act. Passage odds are long before the midterms.

When the cameras already know the school drop-off, the fight moves from product reviews to federal statute. The bill's nickname names one vendor; the text aims at the whole plate-reading class.


Apple just put a harder gate on Mac apps that want to see everything

Setup: TechCrunch reports Apple is adding controls around macOS Full Disk Access — the setting that lets an app read almost everything on the machine, originally so backup tools could work — because AI agents have increased "the risks associated with this level of access." Apple said some developers are using Full Disk Access in ways that could put users at risk "without users' full knowledge and understanding," and that going forward users who "genuinely wish to grant an app this extraordinary level of access" will need "very explicit user action." As AI agents become more capable, Apple said, those risks "will grow substantially."

The Verge and Engadget place the move days after complaints that desktop AI agents encourage Full Disk Access, including disputes around Meta's Muse Mac app. Apple's own Developer News note repeats the same "very explicit user action" language. Apple did not say when the update ships.

Full Disk Access was built for backups. Apple is now saying agents turned that same permission into a privacy problem, and the yes-click is about to get harder to miss. (Friday evening locked this peg for X and never queued it — the filing still stands.)


Prosecutors just charged a Californian over an alleged ~$300 million Nvidia chip pipeline to China

Setup: The Register reports California business owner Greg Lui, 38, was arrested Thursday and charged with allegedly smuggling about $300 million worth of Nvidia hardware to China without proper export licenses. Prosecutors say Lui used Earthmade Computer Inc to order U.S. servers packed with high-powered Nvidia components including A100, H100, RTX 4090, and RTX 5090 GPUs, then shipped them to Malaysia and Singapore for onward movement to China. The scheme allegedly ran from around October 2023 through at least August 12, 2026. Lui allegedly received more than $176 million from two Malaysian transshipment companies and brokered almost $300 million in sales. Charges under export-control, smuggling, and money-laundering counts carry a combined maximum of 50 years as printed.

Engadget frames the arrest against broader due-diligence pressure on Nvidia and Southeast Asia volumes. The spine is export-control enforcement against a physical smuggling pipeline, not a hyperscaler financing story.

The chip shortage is no longer only a cloud-capex problem. When the alleged street price of diverted GPUs clears hundreds of millions, the enforcement story sits next to every board deck about who can legally buy the racks.

Longer cut: Prosecutors charge Greg Lui over alleged $300 million Nvidia chip smuggling to China.

Also noted

  • The Next Web, recounting Business Insider: FieldAI is lining up about $700 million at a $10 billion valuation for software "robot brains" it does not build; term sheet signed, round not closed; revenue/contracts past $135 million across more than 30 customers.
  • The Guardian: Georgia held an emergency elections-board meeting after a Princeton researcher showed public records plus a cheap AI subscription can deanonymize ballots; Springer reported recovering voting order for 1.52 million ballots, or 98.9% of in-person ballots in 114 of 139 counties examined.
  • Gizmodo, recounting Reuters: Dutch eyewear chain Hans Anders is stopping smart-glasses sales in Belgium and the Netherlands over privacy debate.
  • Fortune: GovAI fellows Alan Chan and Sam Manning told Washington labs often run powerful models with internal safeguards off; Chan: "We can't trust them completely to tell us about the safety of models."
  • Friday noon also carried Anthropic IPO prospectus language on government risk, Firmus's ~$7 billion ASX float race, and Amazon's Garman data-center blog / NDA drop (see noon pegs). Companion Fri Brief: Friday AI brief.

Sources


Discover more from Bitter Fool

Subscribe to get the latest posts sent to your email.

Leave a Reply

Your email address will not be published. Required fields are marked *