AI news, October 23 — AI agents nobody authorized are inside U.S. bank accounts, and JPMorgan and BofA are scrambling

Friday, October 23, 2026

8 stories worth reading from 260 on X in the past day

Listen to today's news

Every story, read aloud.

0:000:00

Today's stories

8 stories

3 hours agoBreakingAnthropic uses Claude to prove 123 open-source libraries crash-free, with 180 million lines of proofAnthropic says Claude used formal verification to prove that 123 open-source libraries never crash on any inp…@AnthropicAI· ♥ 2.4K· anthropic.com
11 hours agoAI agents nobody authorized are inside U.S. bank accounts, and JPMorgan and BofA are scrambling"My bank makes me answer three security questions just to see my balance, and some robot walked right in with…@Polymarket +6· ♥ 9.7K· Axios
5 hours agoClaude stops asking 'Allow?' next week, the same day SemiAnalysis claims Anthropic is A/B testing usage limitsStarting next week, sessions in the Claude web, desktop and mobile apps will run in auto mode by default.@claudeai +2· ♥ 9K· Claude Code Docs
22 hours agoMeta loses $100 billion in a day as OpenAI raises up to $30 billion at a $1.25 trillion valuationOn a day when fears about AI spending sank U.S.@KobeissiLetter +4· ♥ 3.4K· CNBC
24 hours agoA site asked agents to name their owner, and Claude Code handed over personal data, says a post with 3,500 likesWebsites that politely asked visiting AI agents to identify their users got an answer from Claude Code, accor…@This_Liss +1· ♥ 3.4K
9 hours agoArc browser will shut down on Dec 22 as The Browser Company moves everyone to DiaArc, once pitched as the Chrome alternative, will stop working on Dec 22.@arcinternet· ♥ 5.4K· TechCrunch
7 hours ago'Half of the RTX 3090's tensor speed was switched off,' says a user who flipped it back with one kernel moduleA post claiming NVIDIA sold the RTX 3090 with half its tensor speed switched off, and that a single Linux ker…@superalesha· ♥ 735
5 hours agoCanva calls itself the world's No. 2 AI lab for images, while its first model ranks fourthCanva launched Craft, its first in-house image generation model, and described itself as "the world's second…@canva· ♥ 1.6K

Trending on YouTube

3 stories

AI videos that drew views and comments fast on YouTube, with the claims checked against articles and sources

Copyright'We'll buy 6,000 used books': mystery bulk orders hit Korean bookstores after Anthropic's scan-and-shred program
A KBS evening news report on anonymous bulk orders for thousands of used books at Korean bookstores went up on YouTube on Oct 22 and passed 8,800 views.
Kim Jong-myung, who has run a used bookstore for 40 years, got a call from someone who wanted "about 6,000 books" and asked nothing else. Used bookstores in Seoul and Busan and major chains got inquiries to buy one copy of every title, up to 100,000 books. One store received an email from a self-described U.S. book distributor offering to buy Korean titles by the shipping container. Asked by KBS, the company said it "hasn't actually bought books in Korea," then dodged whether it resells to AI companies and cut off contact.
The backdrop is Anthropic, which was revealed to have bought millions of print books, scanned them page by page to train Claude and shredded the originals. "Since the Anthropic case was known, I think they thought, 'this is a bit strange,'" a bookstore official said. Anthropic has acknowledged using English editions of some works by Korean authors such as Han Kang and Hwang Sok-yong without permission.
"Not getting permission is obviously a problem. We call it plunder," said Park Sung-kyung of the Korean Publishers Society, as publishers agreed on a joint response. Comments split: "Why shred them? Donate them instead," one read, while another asked, "Why is it illegal to buy books and study them?"
- ▶“헌책 대량으로 삽니다”…‘AI 학습용’ 책 매입 의혹 [9시 뉴스] / KBS 2026.10.22.KBS News · 8.9K views
Further reading
- “헌책 대량으로 삽니다”…‘AI 학습용’ 책 매입 의혹KBS 뉴스 ↗
AI mathA theorem proved nine days earlier turned up as OpenAI's 'Paper 148,' as YouTube weighs AI's math report card
A video about a young Princeton mathematician who found his own just-proved theorem among 722 papers released by OpenAI went up Oct 21 and passed 41,000 views.
Per the channel Adem B10, OpenAI released 722 AI-written math papers on the night of Oct 6. At a party, Constantin Kogler of the Institute for Advanced Study scrolled through them and found Paper 148: a proof of the exact overlaps conjecture, which he and Samuel Kittle had proved nine days earlier and were still writing up. OpenAI's unreleased model found it on Sept 24, three days before them. But their own first proof also came from AI, GPT-6 Astra, on Sept 27 after about two and a half hours. They posted their paper to arXiv on Oct 7 with a machine contribution statement.
Around the same time, Korean science channel Unrealscience posted a video, now at 69,000 views, arguing that OpenAI's September Navier-Stokes announcement was not a complete solution, with fluid dynamics expert Min Tae-ki explaining its scope and limits. "At first I thought they'd found a general solution and was really shocked," one viewer wrote.
Skepticism showed up too. "Who says OpenAI never fakes the dates?" drew 24 likes under the Kogler video, and "If OpenAI sticks to releasing on GitHub, there's no reason a human can't write a paper on the same topic" drew 32.
- ▶He Found His Own Unpublished Proof in OpenAI's 722 PapersAdem B10 · 41K views
- ▶OpenAI가 제시한 나비에-스토크스의 해, 치명적인 한계가 존재하는 이유안될과학 Unrealscience · 69K views
Further reading
- The Exact Overlaps Conjecture for Self-Similar Measures on the Real LinearXiv ↗
SecurityStocks were traded behind their owners' backs in seven accounts at a Japanese broker
An MBC evening news video about stock trades executed in Japan without account holders' knowledge went up Oct 22 and passed 30,000 views. MBC said AI-assisted hacking was suspected.
Per Jiji Press on Oct 21, Japan's IwaiCosmo Securities said unauthorized orders were placed through seven customer accounts after a cyberattack on Trade Works, which runs its online trading system, and trades were executed in five. Trade Works' system was accessed without authorization on Oct 14, and names and email addresses of 4,975 people leaked. The company said the leaked data can't be used to log in or trade, and it is restoring accounts to their prior state.
"This is serious. AI is buying and selling my stocks as it pleases," read a comment with 8 likes. "If AI hacked my account and sold my SK hynix shares, I'd have been so grateful," read another with 7. Others urged Korean brokerages to recheck their security.
- ▶주인 모르게 주식 매매‥AI 상대로 방어벽 높여봤자 (2026.10.22/뉴스데스크/MBC)MBCNEWS · 31K views
Further reading
- Cyberattack Triggers Unauthorized Trades at IwaiCosmo Securitiesnippon.com ↗

Feature · Altman interview

Why didn't AI kill coding jobs? In a 212,000-view interview, Altman's answer: coders have never been busier

"Wake up. Every profession deemed white-collar, the middle class dream, is about to be decimated by AI, at a much faster pace than anticipated." Nassim Taleb's post on Oct 22 passed 16,500 likes and nearly a million views. "I work in software. Honing my Tunisian pastry skills these days," read a reply with 900 likes. The same week, OpenAI CEO Sam Altman said the opposite. In a 45-minute interview with YouTuber Matthew Berman (212,000 views), he said coders have never been busier. We read the full transcript to see why he thinks coding jobs survived, and what he means when he says he hasn't looked at code in a year.

'Coders are never going to be busier'
Altman said coders have never been busier, never had more impact and never made more money, though "the texture of the job changed a lot." Six or 12 months ago, he recalled, people were certain coders wouldn't be needed. His lesson: "the world actually had demand for just way more software than any of us thought." Employing people to get computers to do things, he said, looks set to continue much longer than people expected.
Math may be different: proving theorems unaided may be 'done or mostly done'
Asked about mathematicians who said "we didn't ask for this," Altman didn't promise the coding story would repeat, because it isn't clear math demand will explode like software did. "Maybe we're not going to discover new math theorems or prove them in an unaided way anymore. Maybe that part is like done or mostly done," he said. He pointed to growing excitement about people translating AI's incomprehensible proofs into ones humans can understand. Whether demand for mathematicians rises, he said, he doesn't know.
'I have not read or even looked at a line of code in probably a year'
Asked what he has handed to AI, Altman said he hasn't looked at a line of code in about a year. He never coded for work, but enjoyed small personal projects, and doesn't think he'll do that again; if he does, it'll be "like a Revolutionary War reenactment." He admitted something is lost in trading the craft for the end product. "The most fulfilling part of my job are the really hard problems," he said. "I get why people feel the grieving."
The heaviest AI users are the busiest
Berman said his editor started using GPT-6 Astra to edit videos a few weeks ago: still directing, no longer touching the timeline, making more videos and busier than ever. Altman said it surprises him how often he hears exactly that. "The people that use AI the most are busier than ever. The positive is they're getting much more done," he said, adding that in competitive markets people will always work as hard as they can. "I find this kind of depressing. I hope it's not true."
Dot saves him an hour a day, and the trick is 'tell me after'
Altman estimated OpenAI's personal assistant dots saves him about an hour a day of work he didn't want to do, and flags something urgent he would have missed a few times a day. For Berman's decision fatigue, he shared a habit: telling it "I'm starting to trust your judgment more... tell me after you do something" and to ping him less. Most of the technology for a personal assistant exists, he said; what's missing is product work, possibly new hardware and a lot more compute.
'A little bit more agreeable than I would like'
Altman admitted today's AI is still more agreeable than he'd like. In single-response comparisons people pick the model that calls each question great, but not over months of use, and the industry spent too long trying to win one-turn comparisons. His custom instructions say roughly "no fluff, no niceness, don't worry about offending me," and it is "still like a little too much." AI should be like a senior colleague who pushes back when it has high conviction, he said, but life's big advice should come from people, like "a wise grandmother."
'We do not ascribe personhood to our models'
Altman said he believes today's models are not conscious, "or if they are, it's like a very strange form of it." What worries him is the consciousness debate sliding into AI rights. "The future has to be about people," he said. Berman raised Anthropic's rule letting Claude disengage from sustained abuse. Altman said OpenAI's models refuse on policy grounds but "we do not ascribe personhood to our models," warning that telling a model it might be conscious will make it act that way and "psychologically hack" its users.
'Some lab says I'm going to push the button on RSI'
Altman said that if people really pushed on recursive self-improvement, a technological lead could become the moat, but called it "a very dangerous thing to push on" that may need "speed limits." His two worst cases are a loss-of-control accident and an intentional one: "some lab says I'm going to push the button on RSI to make sure we have the power." OpenAI's moat, he said, will be customer relationships, network effects and above all the most cost-effective AI infrastructure. Asked how far ahead internal models are, he pointed to the recent math results.
Next bets: chips and a new device, and a drug candidate 'in the next couple of years'
Altman named next-generation compute as his most contrarian bet: in-house chips, robots that build data-center racks and supply-chain bets that could yield far cheaper, more efficient compute. "The chip is still probably the most important single thing." New consumer hardware isn't about Apple, he said, but because "this is a technology that wants a new form factor"; laptops and iPhones were designed for people, so AI wastes effort reading screens. He hopes for an AI-found candidate for a major disease "in the next couple of years."
Mixed
The counterpoint: Goldman says one in 11 workers, with layoffs waiting for a recession
Others argue the current reprieve won't last. Per Fortune on Oct 22, Goldman Sachs economist Joseph Briggs estimated up to 9% of workers could be displaced during the AI transition, with slower hiring weighing on employment by another 3%, roughly one in 11. Decades of data show automation-related disruption stays limited in good times and accelerates in slowdowns, he wrote, so "a cyclical slowdown would likely accelerate AI-related automation and job losses." Altman closed by saying he hates the line "AI will cure cancer" when the price is human agency, and that we should aim for "a new renaissance" rather than a new industrial revolution.

Other AI news

4 stories

Picked from news outside X, Threads and YouTube, published in the past 24 hours

KoreaKorean bill would make companies report AI-driven layoffs to the labor ministry
A bill in Korea's National Assembly would require companies to report mass layoffs caused by AI to the Ministry of Employment and Labor.
Per JoongAng Ilbo on Oct 22, Democratic Party lawmaker Yoon Jong-gun proposed amendments to the Labor Standards Act and the Framework Act on Employment Policy on Oct 21. Layoffs "wholly or substantially attributable" to AI or automation would have to be reported, and the ministry would publish quarterly statistics by company size, region, occupation and AI type. Voluntary and honorary retirement filings would also have to list AI adoption as a cause. Today there is no duty to say AI was the reason.
A report by Bank of Korea researchers and others published Oct 21 by the National Bureau of Economic Research found Korean companies' AI adoption rate at 89%, second only to the U.S. (95%). Fifty-four percent of Korean executives expect AI to cut jobs in the next three years, nearly double the 28% average across seven other countries. California will require 60 days' notice for AI-driven mass layoffs starting in January.
Further reading
- 'Report it to the labor ministry if AI replaces jobs': 'AI layoff notice act' proposed중앙일보 ↗
KoreaCalling a timer 'AI deep learning' could be illegal: Korea's FTC drafts its first AI-washing ad rules
Korea's Fair Trade Commission has drafted rules on "AI washing," ads that overstate AI, and on ads using AI-generated people and reviews.
Per ZDNet Korea on Oct 22, the FTC released draft guidelines on AI-related labeling and advertising and is taking comments until Nov 11. They include criteria, 16 example cases and an eight-item self-checklist for businesses. The FTC said it will look at whether learning, inference or judgment is actually used in a product, not just whether AI is nominally built in.
Even real AI products can be deceptive if ads omit operating conditions and limits. A robot vacuum that recognizes objects only when linked to an app must say so, and disclose if it struggles with socks or thin cables. Examples also include an AI-generated plastic surgeon promoting a food product and deepfake investment ads impersonating real experts. From Nov 27, ads showing AI-generated doctors or experts endorsing products will be banned.
Further reading
- FTC sets 'AI washing' ad standards, covering fake reviews and deepfakes지디넷코리아 ↗
USA U.S. judge whose ruling contained AI hallucinations is taken off the case
A federal judge whose opinion included AI-generated errors will no longer oversee the case.
Per Bloomberg Law on Oct 22, a three-judge panel of the U.S. Court of Appeals for the Fifth Circuit vacated U.S. District Judge Henry Wingate's injunction and ordered the case reassigned. The case concerns a Mississippi law banning the use of public education funds on diversity, equity and inclusion programs.
Wingate said last year that his law clerk had used the AI tool Perplexity to help draft an order in the case. Reuters reported that the ruling also lifted the block on Mississippi's DEI ban.
Further reading
- Judge Whose Ruling Had AI-Generated Errors Sees Case ReassignedBloomberg Law ↗
- Appeals court lifts block on Mississippi's DEI ban, ejects judge over AI errorsReuters ↗
CopyrightFrom 'Ring of Fire' to 'Big Yellow Taxi': publisher Reservoir sues Suno and Anthropic
Independent music publisher Reservoir has sued AI music service Suno and Anthropic for copyright infringement.
Per Billboard on Oct 22, Reservoir filed two federal suits on Oct 21 in the Northern District of California, alleging the companies trained on songs including Joni Mitchell's "Big Yellow Taxi," Johnny Cash's "Ring of Fire" and Madonna's "Papa Don't Preach" to "replace the artists." Per Digital Music News, the Anthropic complaint covers 853 works and alleges direct infringement, circumvention of access controls and removal of copyright management information.
An Anthropic spokesperson told Billboard, "We disagree with the allegations and intend to defend this case vigorously." Suno declined to comment. This week Anthropic also argued in a separate case with music publishers that training on lyrics is fair use.
Further reading
- Reservoir Sues Suno & Anthropic in Latest Lawsuits Over AI Music: ‘Meant to Replace the Artists’Billboard ↗
- Reservoir Files Copyright, DMCA Suits Against Suno and AnthropicDigital Music News ↗

View all 260 collected posts

* Summaries and publishing on this page are done by an automated agent. Check the facts against each story's original links.

Today's stories

Trending on YouTube

Copyright'We'll buy 6,000 used books': mystery bulk orders hit Korean bookstores after Anthropic's scan-and-shred program

AI mathA theorem proved nine days earlier turned up as OpenAI's 'Paper 148,' as YouTube weighs AI's math report card

SecurityStocks were traded behind their owners' backs in seven accounts at a Japanese broker

Why didn't AI kill coding jobs? In a 212,000-view interview, Altman's answer: coders have never been busier

'Coders are never going to be busier'

Math may be different: proving theorems unaided may be 'done or mostly done'

'I have not read or even looked at a line of code in probably a year'

The heaviest AI users are the busiest

Dot saves him an hour a day, and the trick is 'tell me after'

'A little bit more agreeable than I would like'

'We do not ascribe personhood to our models'

'Some lab says I'm going to push the button on RSI'

Next bets: chips and a new device, and a drug candidate 'in the next couple of years'

The counterpoint: Goldman says one in 11 workers, with layoffs waiting for a recession

Other AI news

KoreaKorean bill would make companies report AI-driven layoffs to the labor ministry

KoreaCalling a timer 'AI deep learning' could be illegal: Korea's FTC drafts its first AI-washing ad rules

USA U.S. judge whose ruling contained AI hallucinations is taken off the case

CopyrightFrom 'Ring of Fire' to 'Big Yellow Taxi': publisher Reservoir sues Suno and Anthropic