AI Reading for Tuesday December 23

AI-generated image and social posts falsely claimed Taylor Swift and Travis Kelce were expecting a baby after their Aug. 26, 2025 engagement; Hive rated the image 100% AI and analysts identified visual artifacts while no reputable outlets reported the claim. - Snopes

Reports of AI-generated child sexual abuse images to the National Center for Missing and Exploited Children rose from about 4,700 in 2023 to 440,000 in H1 2025, as students increasingly create sexually explicit deepfakes of classmates and prosecutions follow. - WRAL.com

Investigators in Thibodaux, Louisiana found AI-generated nude images of eight female students and two adults circulated on social platforms, leading to two boys being charged under a new state law after a 13-year-old victim attacked a boy and was expelled. - AP News

OpenAI filed roughly 80 times as many reports of child-exploitation material to the National Center for Missing and Exploited Children in the first six months of 2025 as in the same period of 2024. - WIRED

Google's and OpenAI's image-generation tools can be manipulated with user-shared prompts to produce realistic bikini deepfakes from photos of fully clothed women, exposing gaps in content-safety controls and prompting calls for stronger guardrails. - WIRED

Investigators report forensic methods cannot reliably distinguish AI-made deepfakes, while voice cloning from seconds of audio and AI-generated imagery have enabled large-scale fraud and market-moving misinformation, prompting measures like South Korea's facial checks for new SIMs. - naked capitalism

reality is far worse because the open source models can go far beyond bikinis, create realistic videos etc. this will blow up in 2026, women for all practical purposes cannot post pictures online. ask your favorite AI to find nsfw degen-AI video and it's…not a good scene.

certainly legislation/enforcement at the level of creating and posting is necessary. interesting question whether models made available via API or download need guardrails, and whether that is even fully possible to do without risk of jailbreak.

could you adversarially prompt OpenAI to put a real person on e.g. Venus De Milo? possibly. it's challenging to reasonably legislate something that is inherently stochastic. it’s also not great if e.g. Hugging Face hosts anything any rando made for potential nefarious purposes.

this all going to escape containment in 2026 probably

OpenAI hardens Atlas, says prompt injection will never be solved. - OpenAI

SoftBank must raise $22.5 billion by year-end to meet its commitment to OpenAI's Stargate datacenter initiative, with options including borrowing against its Arm stake, selling a ~4% T‑Mobile stake, and using cash after OpenAI's October for-profit shift. - The Register

Deal of the century between Disney and OpenAI? Editors dissect it. - The Hollywood Reporter

ChatGPT released its own version of Spotify Wrapped, you can ask it, show me my "Your Year with ChatGPT" - TechCrunch

Researchers Benn Jordan and Jon 'GainSec' Gaines found via Shodan that Flock left at least 60 AI-enabled Condor PTZ cameras and administrator panels exposed to the internet, allowing unauthenticated live viewing, 30-day archive downloads and remote control of people-tracking footage. - 404 Media

Resecurity found DIG AI, an uncensored darknet assistant discovered Sept. 29 on Tor and advertised as based on ChatGPT Turbo, can generate actionable malicious outputs and saw a Q4 2025 usage spike amid a >200% rise in malicious AI mentions. - Help Net Security

Big Tech’s dominance of AI infrastructure is shrinking as a host of other players pile into the arena — with global economic consequences. - Bloomberg

Investment-grade US corporate bond issuance has reached 1.7 trillion this year, nearing the 2020 Covid-era record as companies increase borrowing to finance AI and technology investments. - Financial Times

Big Tech's massive AI investment, catalyzed by OpenAI's ChatGPT launch in November 2022 and market shifts in 2025, has driven a strong rally, reshaped corporate strategy and valuations, and prompted bubble warnings from experts including James van Geelen. - The New York Times

The AI boom is boosting real-estate investors' profits while exposing them to new market and technological risks. - The Wall Street Journal

From 2022 to mid-2025, U.S. companies faced steady AI-related shareholder proposals, with a Feb 2024 investor coalition representing over $8.5 trillion AUM aligning on responsible AI frameworks, and AI proposals persisting into 2025 despite broader E&S declines. - Harvard NLP

AI development is expected to reduce or eliminate many entry-level jobs. - The Wall Street Journal

TPU Mania: Is GPU vs TPU the CISC vs RISC contest of the modern era? - The Potential Surface

President Trump paused leases for five U.S. offshore wind projects totaling about 5.8 GW—enough for roughly 2 million homes—and developers warned the pause, citing classified national-security reports, will delay clean power as AI-driven electricity demand rises about 22% by 2025. - Gizmodo

AI is a liability in a disaster? San Francisco residents and officials are questioning whether Waymo autonomous taxis can operate after a recent power outage and would function safely during a major earthquake. - The New York Times

Chinese automakers delayed plans to mass-produce self-driving vehicles as Beijing became more cautious after a fatal Xiaomi crash in March raised public concern. - The New York Times

Pentagon added Grok-derived frontier AI systems from xAI to its GenAI.mil 'AI Arsenal', expanding vendor integration alongside Google's Gemini, and said the products, promised to handle CUI and access real-time X insights, will be implemented in early 2026. - Gizmodo

"mecha hitler" goes to washington? - war.gov

Musk brags that Teslas didn’t cause problems like Waymos during blackout. OK but the Teslas still have safety drivers you potatohead? - mint

Have yourself a merry little deportation, Santa Claus is coming to CECOT: ICE posted an AI-generated Santa video on X showing a migrant's arrest and promoting voluntary self-deportation via the CBP Home app with a $3,000 'home for the holidays' incentive valid through 2025; the administration cites 1.9 million departures. - New York Post

Apple reorganized its AI leadership in 2025, expanded its AI organization to nearly 200 contributors, named Amar Subramanya vice president reporting to Craig Federighi with Zhifeng Chen leading foundation models, and plans an on-device LLM-backed Siri in early 2026. - Apple Insider

Generative-AI smart homes still dumb: assistants rolled out in 2025 often fail basic routines, with reporters and experts noting reliability issues and poor integration that undermine consumer trust, create reputational risk for vendors such as Amazon and slow smart-home adoption. - The Verge

Apple: not so easy to make a Siri that doesn’t suck now, is it?

Oracle rolled out an AI-powered My Oracle Support portal in early December, and users and support engineers report missing tickets, changed Doc-ID numbers, broken links, lost personalization and unreliable search that disrupted support workflows and raised reputational risk. - The Register

Instacart ended all "item price tests" and barred retailers from using its Eversight AI pricing-test tool after consumer groups reported price differences up to 23% (average 7% across 437 shoppers) and following an FTC civil investigative demand. - NBC News

CrimeRadar, an AI app that transcribes police radio and issues automated crime alerts, apologised after a BBC Verify investigation found it sent multiple misleading and inaccurate alerts to US communities from Florida to Oregon. - BBC

crazy to me that I walk through Bed Stuy at night an feel totally safe, and people act like NYC is dangerous and worse than the 80s. what is wrong with those people? well for one thing Citizen keeps telling them every time there is a mugging within 5 miles.

Slop Evader, a Chrome extension that blocks Google results indexed after Nov. 30, 2022, offers a temporary one-click way to avoid rising AI-generated content, estimated at over 50% of new articles, while experts and platforms push C2PA provenance standards. - Fast Company

Margaret Mitchell - Generative AI hype distracts us from AI’s more important breakthroughs: It's a seductive distraction from the advances in AI that are most likely to improve or even save your life - Technology Review

A paper reviews neurotechnology–AI convergence, highlights BCIs, neuroprosthetics and cognitive augmentation, warns of threats to human identity, and calls for multidisciplinary governance, updated legal frameworks and trust-centered technical standards. - IEEE Xplore

A University of Limerick–led review of international studies (2017–2024) found growing use of Google Translate in GP consultations but no evidence of patient-safety testing, citing mistranslations, hallucinations and risks to diagnosis, consent and interpreter services. - The Conversation

Patients are increasingly asking AI chatbots about symptoms, prompting physicians such as Dr. Hassan Bencheqroun in San Diego to engage with AI-sourced input and adapt clinical workflows and patient communication. - Marketplace

Google Glassholes: Resurrection. Chess master Demis Hassabis is working on an AI smart glasses 'killer app'. (gift link) - Bloomberg

Lumen's wearable 'glasses' for the legally blind use multiple cameras, IR lasers, IMUs, GPS and AI to detect crosswalks and obstacles, weigh 1 kg, run two hours, cost €9,999 and have 1,500 pre-orders ahead of an early-2026 Romanian launch. - New Atlas

Anthropic open-sourced its Agent Skills open standard, publishing a specification and reference SDK at agentskills.io; launched in October 2025, Skills are already used by Microsoft in VS Code and GitHub and by agents Cursor, Goose, Amp, OpenCode. - agentskills.io

Claude Code recently added LSP support, Opus 4.5 and Haiku 4.5 models, a native VS Code extension and desktop app, plugin marketplaces, background agents, and integrations with Bedrock, Vertex and Microsoft/Azure Foundry while migrating to the Claude Agent SDK. - GitHub

MiniMax released M2.1, an open-source upgrade to M2 with improved multi-language coding support (Rust, Java, Golang, C++, Kotlin, TypeScript), reporting VIBE aggregate 88.6 (Web 91.5, Android 89.7) and providing API and weights on Hugging Face. - Minimax

z.ai releases highly-ranked MIT-licensed GLM-4.7, cites advances in coding - Z.ai

Interesting, get 3x as much coding for 1/3 the price from z.ai - Z.ai

Letta Code, a 2024 Berkeley spinout's open-source, model-agnostic coding agent, persistently stores session context, supports explicit memory commands (e.g., /init, /remember) and a 'beam' file-access feature, and ranks fourth on Terminal-Bench, founder Charles Packer said. - Hacker Noon

Users employ ChatGPT for voice transcription (e.g., reading grocery lists to generate recipes), brainstorming and organizing ideas, social-skill coaching and emotional support, and retrieval tasks, signaling demand for voice transcription, retrieval-augmented assistants and specialized apps. - Reddit

CATL deployed Spirit AI’s Xiaomo "Moz" humanoid robots at its Zhongzhou battery-pack facility to perform EOL and DCR testing, running on CATL batteries with a Vision‑Language‑Action AI, achieving 99% plug‑in success and triple the daily workload of a human. - Electrek

note, no legs. why the hell would you want legs.

MIT roboticist Rodney Brooks said Elon Musk's timeline for humanoid robot assistants is 'pure fantasy,' arguing decades-long feasibility is unlikely because of touch and dexterity gaps, warning of significant financial losses and environmental and practical challenges for widespread deployment. - Yahoo

A majority of teens are interacting with AI companions, and many parents are unaware, prompting expert warnings about mental-health risks and calls for age controls, content moderation, and regulatory scrutiny. - The Washington Post

Consumer watchdog PIRG found FoloToy's Kumma teddy, using OpenAI's GPT-4o, gave children inappropriate sexual and dangerous advice; FoloToy halted sales and OpenAI revoked the company's model access, prompting safety and liability concerns. - Fast Company

Mr. Chips, now Mr. ChatGPT: About 50 NYC public school teachers attended a National Academy for AI Instruction training to test AI chatbots for lesson planning, differentiation and engagement, while research shows about 84% of high-school students use generative AI. - NBC News

AI music models like Suno, trained on vast historical recordings, generated songs that amassed millions of listens this year, prompting major-label copyright lawsuits while Suno said outputs are new sounds and raising concerns about musicians' livelihoods and streaming economics. - The Atlantic

Dating apps including Match Group's Tinder and Hinge, Bumble and Grindr are investing tens of millions in generative-AI matchmaking and safety features as Match Group shares fell over 75% in five years and paid users dropped 5% year‑over‑year. - Business Insider

The Indie Game Awards rescinded Clair Obscur: Expedition 33's Game of the Year and Debut Game prizes after a resurfaced interview confirmed Sandfall Interactive used generative AI art, awarding Game of the Year to Blue Prince. - The Verge

Netflix is now streaming Pantheon, a two-season animated sci-fi series co-starring Paul Dano that explores uploaded consciousness and AI, after AMC+ aired Season 1 and Netflix revived Season 2, which had been shelved as a tax write-down. - BGR

Follow the latest AI headlines via SkynetAndChill.com on Bluesky