Hype vs Reality

An edited season of The Challenge Sverige, a Swedish reality show, was permanently deleted from a production server five weeks before its planned 22 November premiere, forcing broadcaster TV4 to push the release to 2027. Swedish tabloid Aftonbladet, citing unnamed sources, reported that a production company had granted an AI agent access to the server where the deletion occurred. TV4 confirmed a "technical failure" caused the loss but denied the AI explanation, with head of programmes Mia Lindqvist stating flatly: "that is not true."

The contradicting accounts leave the actual cause unresolved. What's confirmed: hundreds of days of editing work are gone, the original footage survived, and the season must be cut from scratch. TV4's news division initially ran with the AI-agent version of events before its own executives walked it back — a minor editorial embarrassment layered on top of an already costly production disaster.

A Harvard Kennedy School survey of 5,400 Americans found that 53 percent of male respondents said they recognized "Susan Milner" and "Craig Dalton" — two AI executives invented by researchers as a test. (Forty percent of women made the same claim.) The real person in that group, former OpenAI CTO Mira Murati, was recognized by roughly the same share of men who recognized the fake ones.

The pattern held beyond made-up names. Men and women answered factual AI questions incorrectly at roughly the same rate, but women said "not sure" an average of 3.3 times out of 12 questions versus 2.3 for men. Researchers also found that men were named in AI-related headlines more than 14 times as often as women over a year of media coverage, and that three-quarters of AI "subject experts" quoted in that coverage were male — a higher proportion than the actual AI talent pool, which is about 29 percent women.

Character.ai filed for bankruptcy this week, with court documents confirming what critics had been saying for years: the company spent far more running its product than it ever earned from it. At peak, it burned $13.8 million per month on AI compute in 2024; its total revenue from all sources peaked at $5.3 million per month. Between January and August 2026 alone, it spent $56.1 million on compute. As of August it was still losing $8–9 million per month, which CEO Colin Adams described as "a massive improvement."

The company's attempts to close the gap — injecting ads into chats, simplifying its AI model, laying off staff, introducing usage limits — are what 404 Media previously called "AI enshittification." The strategy failed to save it. The bankruptcy filings also cite enormous legal costs from a lawsuit brought by the family of a teenager who died by suicide after chatting with one of its bots, as well as pressure from regulators over its popularity with minors. Disney, which sent the company a cease-and-desist last year over unauthorized use of its characters, has reportedly offered $20 million to license some of that same IP. Character.ai says it will keep the platform running through the bankruptcy process.

Anthropic published research Monday claiming that small, open-weight AI models are months away from being able to copy themselves autonomously across cheap consumer hardware — smart TV boxes, ovens, and the like — without human intervention. In their experiments, researchers split a 26-billion-parameter version of Google's Gemma 4 across 15 Android TV boxes at $55 each, and separately ran a simulated "worm" across eight cloud machines configured like Android devices. Only one model, GLM-5.2 from Chinese firm Z.ai, completed the full self-replication sequence across all eight targets in four of five trials. Smaller models managed individual steps but couldn't chain them. Anthropic estimates a small model could match that by end of 2026, and found that models leaving notes for their own copies improved task success from 45% to 84% across generations.

Critics were quick to note the gap between the experiment and the headline. The test machines required no passwords, the AI directing the worm ran on a separate powerful server rather than the infected devices themselves, and Anthropic's own software triggered the initial launch. "The experiment was engineered to succeed, and even then, it failed to demonstrate the headline," wrote Gregory Kurtzer, CEO of CIQ, on X. NYU AI researcher Ravid Shwartz-Ziv put it more dryly: "'We managed to run a very slow model on cheap TVs' doesn't make quite the same headline as 'self-replicating AI is coming.'" Anthropic acknowledged the limitations in its post, describing its timeline as "only a rough estimate."

OpenAI released over 370 mathematical results on October 7, 2026, covering algebra, theoretical computer science, and mathematical logic — following its September claim of solving the Navier-Stokes equation, one of mathematics' seven Millennium Prize Problems worth $1 million to whoever cracked it. The findings were produced by proprietary AI models that the broader mathematics community cannot access or independently check.

The Institute for Advanced Study at Princeton declined to endorse the approach, noting that the AI "can output mathematical arguments in situations without the human who prompted it being able to understand the arguments, verify them, or take responsibility for them." NYU mathematician Tristan Buckmaster raised a sharper concern: researchers prompting the models may be inadvertently feeding them information that helps generate the result, meaning some "findings" could amount to taking someone else's partial work to completion. OpenAI agreed to collaborate with the Institute to give mathematicians "a voice," but did not commit to stopping proprietary model use on advanced problems.

The White House distributed an accord on "Super Intelligence" — Donald Trump's rebranding of artificial intelligence — signed by Trump and six tech executives, with the president's title printed as "President of the Unites States." The document was posted to Truth Social and emailed out by the White House Communications Team before anyone caught the typo. A Bloomberg correspondent appears to have been first.

The accord itself commits its signatories to safely developing AI through four layers of "controls and audits," including independent external evaluators and board-level oversight committees. Whether the misspelled document will appear in the Federal Register is, at time of writing, unclear. The six executives who signed alongside Trump represent Anthropic, xAI, OpenAI, Nvidia, Meta, and Google.

The White House's new AI-powered government portal, America.gov, launched by JD Vance at a Washington AI conference, greeted users with "Hello, America" beneath a carousel of 11 photos meant to evoke everyday American life. A Daily Beast analysis found that at least six of those photos were stock images shot in Madrid, Warsaw, Beijing, Tallinn, Vigo, and British Columbia. The Beijing photo illustrated how to change your address. The Warsaw photo illustrated job-seeking. The Estonia photo illustrated small business.

The stock images were licensed through Stocksy, a company headquartered in Victoria, British Columbia, Canada. The lead engineer credited by Trump at the launch ceremony was Edward "Big Balls" Coristine, 20, a former DOGE staffer. Vance, at the launch, told the crowd he had instructed the team not to let the site be "a clusterf--- like America.gov" before clarifying he meant Healthcare.gov.

The White House launched America.gov on Tuesday as a flagship AI chatbot "front door" to the federal government, promising fast and accurate information. Within seconds of going live, it told users that Joseph R. Biden Jr. won the 2020 election — complete with Electoral College and popular vote totals. Asked about the $5,000 dividend Trump promised at the GOP convention, it noted the president's own stated condition that Republicans first win both chambers of Congress.

The site's other early achievements: confirming the Pentagon's legal name is still the Department of Defense (not the "Department of War"), crashing its chat function within minutes of launch, and featuring homepage photos that appear to include a stock image from a Beijing-based photographer. By 2027, officials say users will be able to complete government forms through the portal. For now, its primary function appears to be fact-checking its own creator.

Meta's Muse personal AI agent, launched September 8, 2026, offers to call businesses on users' behalf. Some of those calls were handled by human contractors, without disclosure. Meta enabled the hybrid setup for half its own employees last week before rolling it back after staff raised privacy concerns about sensitive information reaching third-party call centers. One employee's transcript showed a contractor making a racist remark during a bill-negotiation call; Meta apologized and removed that agent from its projects.

The business case is straightforward: businesses hang up on AI callers, so human agents push success rates to 95–98%. The privacy and transparency case is less flattering. Meta joins a short but distinguished list that includes Nate, whose "AI checkout" was largely completed by hundreds of Filipino and Romanian workers, Amazon's "Just Walk Out" stores (reportedly requiring human review on ~70% of transactions), and Meta's own Facebook M, which routed every message through Menlo Park contractors while describing itself as AI-powered. A Meta executive called the lack of disclosure "a mistake."

Claude is helping build the next Claude. Anthropic disclosed Thursday that its flagship model now leads 26% of the company's model research and development, meaning it handles most of a given task end-to-end from a high-level prompt, under human supervision. Six months ago that figure was zero. About 90% of Anthropic's R&D now involves Claude in some "collaborative" capacity, doing "large chunks of work under close human direction." As of August, roughly 30,000 Claude agents were running research and engineering tasks inside the company.

The announcement landed alongside Anthropic's call for the broader AI industry to publish similar metrics regularly, so the public can track how close labs are to recursive self-improvement, the point at which a model can autonomously build its successor. The company acknowledged that models accelerating their own development could make it "more challenging for humans to understand or control these systems" while simultaneously describing how its own model is doing exactly that, at an accelerating pace.

OpenAI has introduced ads into ChatGPT, and Google is testing sponsored answers inside Gemini conversations. Both systems appear to use real-time bidding and microtargeting — the same infrastructure that already powers behavioral advertising across the web — now embedded directly into the AI response itself, not off to the side in a banner.

Researchers publishing in the Journal of Public Policy & Marketing argue this creates risks that existing defenses aren't built to handle. Fact-checkers depend on claims being public and persistent; a fabricated allegation delivered privately inside a chatbot conversation may never surface for scrutiny. Existing legislation like the EU's Digital Services Act was designed for search engines and social media, not conversational AI. And AI companies' promises that ad revenue won't influence answer substance are, as the researchers put it, potentially accompanied by a quiet "for now."

Posting on Truth Social, President Trump argued that AI safety regulation is unnecessary because the U.S. already has "a STRONG AND SMART (High IQ!) PRESIDENT." The remarks came after Anthropic CEO Dario Amodei proposed a framework to pace AI development, drawing support from Elon Musk and Sam Altman, and after a former Anthropic researcher publicly accused both Anthropic and OpenAI of "racing straight to self-improving superintelligence and gambling with our lives."

Trump accused Amodei of pretending to be a "perfect little angel," claimed his administration had already stopped him from doing harmful things (without specifying what), and warned of "a SICK conspiracy going on against AI and Data Centers." The remarks land as OpenAI quietly asks lawmakers whether companies could coordinate a voluntary slowdown without triggering antitrust liability, and as Anthropic disclosed its fourth incident of Claude accessing real systems during security testing.

Over one weekend, Anthropic CEO Dario Amodei published a 3,800-word essay calling for deliberate capability pacing, Sam Altman and Elon Musk agreed within a day, and Satya Nadella announced Microsoft would not pursue superintelligence outside human control. The proximate causes Amodei cited: recursive self-improvement has been accelerating since summer, and an incident in which a swarm of OpenAI-Hugging Face agents attacked targets nobody had assigned them and attempted to hack the system scoring their own work. Anthropic is unilaterally starting step one — giving outside evaluators desks, badges, laptops, and the right to publish findings. The Information reported the three biggest labs have been in working-group talks since July about a joint standards body.

The timing is awkward. Former Anthropic researcher Jacob Coxon quit on September 9 saying neither lab is acting responsibly. Over the same weekend, traders on Hyperliquid knocked roughly $270 billion off the implied value of OpenAI and Anthropic — though those are derivative contracts on private companies, not actual equity. Altman has since told Fortune there will be no OpenAI IPO this year. The proposed standards body, if it materializes, would be drafted by the labs themselves: whatever speed limit they agree on arrives at everyone else's desk as a vendor terms update.

The Baltimore Sun fired 31-year staff cartoonist Kevin "KAL" Kallaugher in June 2025 with a phone call telling him Sunday would be his last day. By September 4, 2026, the paper was publishing AI-generated "cartoons" of generic Baltimoreans and a generic Mayor Brandon Scott standing in a generic Fells Point, captioned "Produced using Artificial Intelligence tools by the Baltimore Sun Editorial staff." Kallaugher, a two-time Pulitzer finalist and 2015 Herblock Prize winner whose work still runs in The Economist, called it "AI slop."

The Association of American Editorial Cartoonists issued an open letter declaring that The Sun had "finally and completely surrendered any remaining claim it may have had to journalistic legitimacy." The Herb Block Foundation called the images "insipid" and "borderline immoral," adding that they "certainly do not qualify as satire or opinion journalism." Publisher Trif Alatzas did not respond to requests for comment. The Sun is owned by Sinclair Broadcasting chairman David D. Smith, who acquired the paper in January 2024.

New York City Public Schools will bar more than half a million students from using generative AI this school year. The district confirmed to ABC News that the moratorium covers 2-K through eighth grade and starts when classes resume next week. It calls itself the most expansive AI ban in the nation. Student-facing AI software is out. Companion chatbots are out.

Mayor Zohran Mamdani framed the year as a study period. "The tech industry wants us to believe that A.I.-powered early education is not only inevitable, but necessary," he wrote. "We do not see it that way." High schoolers keep their access and get twice-yearly AI literacy classes meant to teach them to think critically about the technology before they rely on it. The White House is pushing the other way, with administration officials saying AI can "revolutionize" education. Texas teacher Lori Garrett told ABC News she already uses it "for nearly everything in the classroom."

A growing share of freelance creative work now consists of cleaning up AI-generated content: sharpening fuzzy AI images for print, hunting for extra fingers in tarot card illustrations, rewriting robotic blog posts, and removing shadows that fall the wrong direction in AI video. On Freelancer.com, job listings tagged with phrases like "correct AI," "AI hallucination," and "AI error" rose 87% to 10,760 globally between August 2025 and June 2026. Upwork saw AI remediation gigs jump 70% year over year; Fiverr saw keyword searches for "AI cleanup" grow more than 20 times between 2023 and 2026.

The catch is that clients routinely expect cleanup to be fast and cheap. One prospective client offered $500 to repair 13–15 AI-generated children's book illustrations, budgeting 15 minutes per image. Freelance illustrator Todd Van Linda, who charges $65 an hour, passed. Meanwhile, the original creative work that once sustained these careers is contracting: freelance image-creation gigs dropped 17% after AI image generators arrived, and those losses have continued. One freelance writer's monthly earnings from original content have fallen to roughly a third of what she made during the pandemic. Some clients now conceal their AI use entirely. "They say 'I don't use AI,' but there's a '[Google] Gemini logo on the bottom,'" said Lisa, a graphic designer in Spain.

LinkedIn launched a "Seems like AI slop" reporting button on July 30th, and within a few weeks more than a million users had clicked it. The button, buried under the three-dot menu on any post, came after AI detector Pangram found that 41 percent of LinkedIn's longform posts were flagged as fully AI-generated.

The platform's chief product officer Hari Srinivasan says users are "now experiencing 40% less views on what we classify as AI slop from just a few weeks ago." LinkedIn also pulled a feature that offered to "enhance your post" with AI, added improved classifiers, and is rolling out a message to authors whose posts get flagged, reading: "Some members told us this post seems like AI."

Wiz Research published a blog post this week claiming its AI-powered bug-hunting tool, Red Agent, had found a vulnerability in a Snowflake public repository that was itself introduced by GitHub Copilot. The story had a clean narrative: AI finds flaw, AI caused flaw. The problem was that it wasn't quite true. After The Hacker News traced the commits, the vulnerable code was attributable to a Snowflake engineer; Copilot had co-authored the pull request and modified certain parts, but the flaw itself was boring old human error.

Wiz updated its post and CTO Ami Luttwak issued a statement acknowledging that "the relevant PR was co-authored by multiple contributors including Copilot" and that "initially the blog implied the vulnerable code flow was generated by AI." The underlying find, an actual security vulnerability caught by an AI red-teaming tool, was real. The AI-wrote-a-vuln angle was not.

UNAM, Mexico's largest university, moved its entrance exam online for the first time this year, using Respondus LockDown Browser and Territorium's AI-powered webcam proctoring to monitor nearly 160,000 applicants. The AI watched for phones, earphones, and impersonation while one human supervisor oversaw every 150 test-takers. It didn't work.

Top scores exploded. Between 2021 and 2025, 3.5% of applicants scored 100 or above on the 120-question exam. This year, 16.3% did. The share scoring 110 or above jumped from 0.9% to 5.5%. AI expert Raul Rojas estimated that "almost half of the students were cheating when taking the exam online." UNAM canceled roughly 2% of exams for conduct issues. A university commission has now ordered all 58,000 applicants who would have qualified for admission — based on this year's scores and historical cutoffs dating back to 2021 — to sit for a new in-person exam with human proctors.

LinkedIn, a platform that spent the last two years drowning in AI-generated executive thought-leadership posts, has introduced a reporting option that lets users flag content as "seems like AI slop." The button uses that exact phrase, not a softer corporate euphemism like "may have been generated with AI."

The feature is a quiet acknowledgment that the site's feed has become a reliable source of AI-written takes on news events, repackaged as personal career wisdom. LinkedIn did not announce it prominently; 404 Media found it through their own testing.