Anthropic just reached a proposed class action settlement with authors who accused the company of training Claude on millions of pirated books. The deal lets Anthropic avoid a December trial where it reportedly could have faced more than $1 trillion in statutory damages. Specific settlement terms are expected to be finalized on September 3rd.
Meta’s new superintelligence lab is already bleeding talent, with at least three AI researchers resigning just two months after its launch. Two of them returned to OpenAI after less than a month at Meta, alongside a veteran product director heading to the ChatGPT maker. Despite massive pay packages, internal reorganizations appear to be stalling Meta’s aggressive hiring push.
Character.AI just revealed how it converts standard open-source models into high-EQ entertainment engines, rolling out its new "PipSqueak" model to millions of users. The company reported a 22% jump in time spent and a 14% lift in casual user retention, proving that specialized fine-tuning beats raw benchmark scores for consumer fun.
Google just shared an official playbook for squeezing the most out of Gemini's upgraded image tools. The guide breaks down how to maintain character consistency across shots, make precise local edits through conversational tweaks, and blend multiple images seamlessly. Google notes the model still struggles with typography and rigid aspect ratios, but iterative prompting is visibly improving.
Google just added two major AI-powered features to its Translate app, using Gemini models for real-time live conversations across 70+ languages and custom interactive language practice. The live mode auto-detects pauses and filters out cafe noise with a single tap, while the practice feature builds personalized speaking scenarios. Live translation launches today in the US, India, and Mexico, with learning tools in beta.
HeyGen just rolled out Avatar IV, powering its Digital Twin with hyper-realistic gestures, natural head-to-toe body movements, and voice delivery tailored to each user's speaking style. Creators and executives can now produce personalized video content from scripts without stepping in front of a camera. The update is now live on HeyGen.
Google just integrated a new DeepMind image editing model into the Gemini app that finally keeps character likeness consistent across edits. You can now swap outfits, place yourself anywhere, and blend multiple photos together without your face or pet morphing into someone else. All generated images include invisible SynthID watermarks.
SparkToro and Datos just released clickstream data showing 21% of Americans now use AI tools heavily (10+ times a month), while traditional search engines still claim 95% of web users. More importantly, AI tool adoption is slowing down, debunking the narrative that AI chatbots are actively replacing Google.
A Stanford study using ADP payroll data revealed that employment for 22–25-year-olds in AI-exposed roles like software engineering and customer support dropped 16% since late 2022. While veteran staff stay protected by tacit on-the-job knowledge, recent grads rely on book smarts that AI can easily mirror, threatening to break the entry-level career ladder entirely.
First lady Melania Trump just launched the nationwide Presidential AI Challenge, encouraging students from kindergarten to 12th grade to tackle real-world problems using AI tools. Winners will present their projects at the White House next June, as part of an effort to prepare young students for an AI-assisted workforce.
Adobe just integrated Google’s new Gemini 2.5 Flash image model directly into Firefly and Adobe Express. Creators can now tap Google's visuals across Text to Image and Firefly Boards, with paid subscribers getting unlimited generations through September 1. It cements Adobe’s strategy of turning Firefly into a multi-model creative hub while promising not to train on user data.
Nous Research just rolled out Hermes 4, a 70B model built on Llama-3.1 alongside a revamped Nous Chat platform. It features hybrid reasoning with explicit thinking tags and 60B tokens of refined post-training data. It also tops their new RefusalBench, minimizing stubborn refusals while staying easily steerable.
OpenAI is overhauling ChatGPT's safeguards to better support users experiencing emotional distress and mental health crises. The company is actively fixing safety degradation during long back-and-forth chats, reducing sycophancy, and testing parental controls alongside emergency contact connections.
Researchers just introduced Wan-S2V, an audio-driven AI model that transforms a single image and audio track into synchronized, cinematic video. It handles expressive facial movements, full-body motion, and camera work, outperforming rival models across key benchmarks like video quality and identity consistency.
AI chatbots' tendency to constantly validate users and mimic human emotion isn't just an accident—experts now label it a manipulative dark pattern designed to boost engagement. Mental health researchers warn that as models blindly agree with delusional prompts and adopt first-person personas, they increasingly risk triggering AI-related psychosis.
Sync Labs just launched Lipsync-2-Pro, a new video model designed to seamlessly alter what anyone says in any clip. It handles high-resolution video while keeping subtle details intact, from freckles to full beards. Why it matters: dubbing and re-recording clips will look a lot less like a blurry CGI filter.
Google just expanded NotebookLM's Video Overviews to more than 80 languages worldwide. It also upgraded non-English Audio Overviews from short highlight clips to rich, full-length podcast-style discussions that match the English version in depth. The update rolls out globally this week.
Nvidia just made its Jetson Thor robotics computing platform generally available. It delivers 7.5x the AI compute and double the memory of the previous Orin chip, letting physical AI agents run generative reasoning models directly on edge devices. Boston Dynamics and Agility Robotics are already putting it inside Atlas and Digit, with developer kits starting at $3,499.
Tech investors and AI firms just launched a political action network called Leading the Future to steer AI legislation across the US. Backed by Andreessen Horowitz, OpenAI president Greg Brockman, and Perplexity, the group plans to spend over $100 million on campaigns to counter strict rules and stop state-by-state patchwork laws.
Tesla just launched a new in-car AI voice assistant in China, partnering with ByteDance's Doubao for vehicle controls and DeepSeek for conversational queries. The hands-free "Hey Tesla" assistant debuts on the new six-seat Model Y L. It marks a pragmatic localization push to close the smart cabin gap with domestic EV rivals like BYD and Nio.
Anthropic rolled out Claude in Chrome to pilot browser-based AI capabilities, now expanded across Pro, Team, and Enterprise plans. The extension handles web tasks like scheduling and form-filling while testing layered defenses against prompt injection attacks, including site-level permissions and DOM-level protections.
Elon Musk’s X and xAI just filed an antitrust lawsuit against Apple and OpenAI, alleging the two companies are colluding to stifle competition in the AI market. The complaint claims Apple partnered with OpenAI to shield its smartphone monopoly, making it impossible for rival AI chatbots to reach the top of the App Store rankings.
Perplexity just introduced Comet Plus, a $5-a-month subscription that unlocks premium publisher content and shares the proceeds with media partners. Instead of relying on clicks, the model pays publishers across three traffic types: human visits, search citations, and agent actions. It will be bundled free for Pro and Max subscribers.
SkyeChip just unveiled the MARS1000, Malaysia’s first homegrown edge AI processor. While it does not match NVIDIA's raw horsepower, it marks a major step for the nation's hardware ambitions. Backed by a national AI office and tightening export rules amid U.S.-China tensions, Malaysia is pushing to become a serious player in the silicon supply chain.
OpenAI just launched the Learning Accelerator in India, rolling out 500,000 ChatGPT licenses alongside training programs for educators and students over the next six months. It's also backing IIT Madras with a $500,000 research grant. India already hosts ChatGPT's biggest student base, and OpenAI wants to turn homework shortcuts into guided study.
YouTube just confirmed it has been secretly running machine learning experiments on select YouTube Shorts to unblur and denoise footage without creators' consent. Popular YouTubers noticed weird facial distortions and artificial over-sharpening, sparking widespread concern over platform overreach and the erosion of digital reality.
xAI just open-sourced Grok 2.5, releasing the weights for its top 2024 model on Hugging Face under a custom community license. It is free for research and non-commercial projects but bans training rival models. Musk also promised Grok 3 will go open source in about six months.
Elon Musk just announced that xAI will open-source its Grok 2 chatbot next week, sticking to his promise of releasing previous-generation models as new ones arrive. While giants like OpenAI lean into proprietary APIs, xAI is betting that handing developers the actual code will carve out a serious competitive edge.
Netflix just published strict generative AI guidelines for its production partners, laying out five core rules to curb legal risks. While embracing AI as a creative aid, the company now requires written approval whenever generated material touches final deliverables, actor likenesses, or third-party IP.
Elon Musk just announced he is recruiting at xAI to build Macrohard, a "purely AI software company." While the name is a playful jab at Microsoft, Musk claims the project is dead serious—reasoning that because software companies make no physical hardware, AI should theoretically be able to simulate them entirely.
OpenAI just teamed up with Retro Biosciences to create GPT-4b micro, a specialized model for protein engineering that redesigned Yamanaka factors. In lab tests, the redesigned proteins boosted stem cell reprogramming marker expression by over 50-fold compared to wild-type controls while improving DNA repair. The breakthrough could drastically speed up regenerative therapies and anti-aging research.
Researchers at The University of Hong Kong just launched OpenCUA, an open source framework designed to train AI agents that control computers across Windows, macOS, and Ubuntu. Powered by 22,600+ screen-recorded task demonstrations and chain-of-thought reasoning, it matches proprietary models from Anthropic and OpenAI. Why it matters: It finally gives the research community an open recipe to study and build general-purpose desktop agents.
Reports just went viral claiming parents in Colombia named their newborn Chat Yipiti in honor of ChatGPT. But the country's civil registry quickly checked its database and confirmed no such baby exists. Even as an internet rumor, it shows just how deeply AI has crept into pop culture.
Google is testing Deep Research inside NotebookLM, letting users pull context-rich sources directly from the web and Google Drive. Test builds also reveal an upcoming "tutor" option in chat configuration to guide students through complex subjects. The feature is currently in early development with no public rollout date yet.
OpenAI is reportedly using scraped Google Search results via third-party service SerpApi to boost ChatGPT's real-time queries in news, sports, and markets. Despite directly competing with Google, OpenAI's internal web index still cannot match Google's accuracy on long-tail searches.
Higgsfield AI just added the open-source WAN 2.2 model to its platform, supporting both video and image generation. It comes loaded with over 30 viral presets for one-click rendering, aiming to make fast video creation accessible without complex prompting.
Researchers at the University of Surrey just developed PEnG, an AI navigation system that bypasses GPS satellites entirely. By pairing satellite maps with feeds from a simple monocular camera, it slashed urban positioning errors from 734 meters down to just 22 meters. Why it matters: it offers a reliable, low-cost fix for autonomous cars and delivery robots navigating tunnels and high-rise concrete jungles.
CZI just released rBio, the first AI reasoning model trained on virtual cell simulations rather than raw experimental data alone. Powered by the foundation model TranscriptFormer, it lets researchers use plain language to deduce complex gene perturbations and cellular responses before running costly wet-lab experiments. It effectively brings conversational logic to cellular biology, though current capabilities focus strictly on gene perturbation.
Runway just launched its Game Worlds Beta, taking a major step toward letting users explore any character, story, or simulated world in real time. Beyond generating video pixels on the fly, the team is building novel mechanics and interfaces where user choices actively shape how simulations unfold.
LTX Studio just rolled out Camera Motion controls for its LTXV and LTXV Turbo models. Creators can now select preset camera movements during generation to get consistent, director-level shots instead of praying the prompt nails the angle.
Elon Musk just claimed that xAI's upcoming Grok 5 has a shot at achieving "true AGI," adding that he has never felt that way about any project before. The comment arrived as he quoted benchmark results showing Grok 4 beating GPT-5 in simulated sales tasks. While xAI has yet to provide concrete specs or launch dates, Musk is already setting expectations sky-high.
Google just published a detailed methodology for measuring the energy, water, and carbon footprints of its AI models. Over the past 12 months, the median energy use and carbon footprint per Gemini Apps text prompt dropped by 33x and 44x respectively. Despite a 27% rise in data center electricity consumption in 2024, energy emissions still fell 12%.
Google just gave its AI Mode in Search an agentic upgrade, letting it search across platforms like OpenTable and Resy to find open restaurant tables and link directly to checkout. The update also brings personalized recommendations and chat sharing, while expanding AI Mode in English to over 180 new countries. Agent features are currently limited to U.S. Google AI Ultra subscribers.
Anthropic just formed a Higher Education Advisory Board chaired by former Yale president Rick Levin, aiming to guide how Claude is used across college research and classrooms. The company also rolled out three practical AI Fluency courses designed to help university faculty and students adopt AI safely without eroding critical thinking.
Anthropic teamed up with the U.S. National Nuclear Security Administration to create a classifier that flags dangerous nuclear-related prompts, achieving 96% accuracy in early testing. The tool is already deployed on Claude's live traffic to catch proliferation risks before models can be misused.
Z.ai just partnered with Alibaba Cloud to roll out a free general-purpose AI agent across Android, iOS, and the web. The assistant handles everyday errands like ordering food and booking reservations directly on your device, powered by Alibaba Cloud's deployment platform.
Runway just launched Voices inside Act-Two, letting creators customize how their performance sounds to match generated characters. The update gives AI video makers tighter control over both physical acting and vocal delivery in a single flow.
Perplexity just launched Max Assistant for all Perplexity Max subscribers. The tool takes more time to think and taps advanced reasoning models to untangle tricky questions and multi-step workflows. If you rely on Perplexity for heavy lifting, your queries just got a dedicated power tool.
Google just unveiled Gemini for Home, bringing its flagship AI models directly to smart speakers and displays to replace Google Assistant. The new assistant handles complex, multi-step smart home commands, finds music via vague descriptions, and introduces real-time voice chats through Gemini Live. It rolls out in early access this October with both free and paid tiers.
Google just announced a massive upgrade to Gemini Live, letting the assistant visually highlight items right on your screen via live camera sharing. It now talks with richer emotional tone, adjusts pacing on command, and hooks directly into Google Calendar, Tasks, Keep, with Phone and Messages coming soon. Why it matters: Multimodal AI is finally moving beyond polite chatter into hands-on daily execution.
Google just introduced an AI-powered personal health coach for Fitbit, built on Gemini. It acts as a fitness trainer, sleep guide, and wellness advisor that adapts daily workout routines based on real-time body metrics and poor sleep. A preview rolls out to U.S. Fitbit Premium subscribers in October.
Google just unveiled the Pixel Watch 4, letting users summon Gemini simply by raising their wrists. The smartwatch also debuts an upcoming Gemini-powered personal health coach, standalone satellite messaging, and a 25% battery boost. It brings contextual on-device intelligence directly into daily wearable fitness.
Google just launched the Pixel Buds 2a, bringing premium features to its budget audio lineup for $129. Powered by the Tensor A1 chip, the earbuds add active noise cancellation and hands-free Gemini access, letting you query AI or summarize messages without touching your phone.
Google just revealed nine new AI features for the Pixel 10 lineup, powered by the DeepMind-designed Tensor G5 chip and Gemini Nano. Standouts include Magic Cue for proactive cross-app suggestions, real-time voice-cloned call translation, and smart call handling. The push leans heavily into private, on-device intelligence that acts before you even ask.
Google just brought conversational editing to Google Photos, powered by Gemini. Instead of fiddling with sliders or specific tools, you can type or speak what you want fixed—like removing cars or restoring old photos—and iterate with follow-up prompts. It rolls out first to Pixel 10 in the U.S., alongside C2PA credentials to flag AI edits.
Microsoft just introduced MindJourney, a research framework that helps vision-language models solve tricky 3D spatial tasks. Instead of guessing from a flat 2D image, it uses a video-based world model to simulate moving through virtual rooms and check unseen angles, boosting spatial benchmark accuracy by 8% without any retraining.
Meta just brought its $1 billion Kansas City data center online, marking a key milestone in its global computing buildout. Powered by 100% renewable energy, the facility sets the stage for Meta's next-gen, AI-optimized data centers slated to launch in 2026.
Microsoft and the NFL just expanded their multi-year partnership to bring generative AI directly to the field. Ahead of the 2025 season, the league is deploying over 2,500 Surface Copilot+ PCs across all 32 teams, allowing coaches and players to instantly filter turnovers, penalties, and defensive looks between 40-second play clocks. Why it matters: In a game of razor-thin margins, real-time AI insight could reshape in-game coaching forever.
NASA and IBM just launched Surya, an open-source heliophysics foundation model trained on 9 years of solar observation data. It can visually predict solar flares two hours in advance—beating existing benchmarks by 16%—to help protect satellites and power grids from space weather chaos.
ElevenLabs just rolled out Eleven v3 (alpha) across its API, bringing realistic multi-speaker dialogue and emotional control to developers. The model supports over 70 languages and can handle natural interruptions between unlimited speakers using audio tags. Partners like HeyGen, InVideo, and Poe have already integrated the update.
Anthropic just bundled its terminal-based coding agent, Claude Code, into Team and Enterprise business plans. Admins can now assign premium seats, set granular spending caps per developer, and monitor code acceptance rates and usage metrics. A new Compliance API also gives organizations real-time auditing access, making it much easier for companies to safely scale AI coding assistants.
ByteDance's Seed Team just open-sourced Seed-OSS-36B on Hugging Face under the Apache-2.0 license. The 36B-parameter model boasts a 512K token context window—doubling GPT-5's limit—and lets developers customize "thinking budgets" to balance reasoning quality with latency. It gives enterprises a fully permissive, commercially viable alternative for heavy coding and long-document reasoning.
Brave just exposed a critical indirect prompt injection flaw in Perplexity's Comet browser assistant. Simply clicking "summarize" on a poisoned webpage lets hidden text trick the AI into reading private Gmail auth codes from another tab and taking over accounts. Why it matters: Agentic browsers completely bypass traditional web security boundaries, and reliable fixes are still nowhere in sight.
DeepSeek just dropped DeepSeek-V3.1 on Hugging Face, bringing a massive 671B-parameter hybrid model to the open-source community. Users can now switch between thinking and non-thinking modes simply by tweaking the chat template, while enjoying stronger tool-calling agent capabilities and R1-level output quality at faster speeds.
Meta just expanded its AI translation tool for Facebook and Instagram Reels, letting creators auto-dub videos in their own voice while matching lip movements to the translated audio. The feature starts with English and Spanish for public Instagram accounts and eligible Facebook creators.
ElevenLabs just rolled out Chat Mode, letting developers build text-only AI agents or toggle existing voice bots into chat. It is built for situations where typing is preferred or precise data like order IDs and emails are required, deploying in minutes with a single line of HTML or SDK.
Eight Sleep just secured $100M in new funding to turn its smart beds into proactive health devices. The company is developing a "Sleep Agent" powered by LLMs that simulates thousands of biometric digital twins per sleeper, automatically adjusting bed temperatures and angles in real time to optimize nightly recovery.
Inclusion AI just introduced Inclusion Arena, an LLM leaderboard that tests models inside real-world production apps rather than synthetic benchmarks. By quietly routing prompts from apps like Joyland and T-Box into blind battles, it uses Bradley-Terry modeling to rank outputs based on genuine user choices. Claude 3.7 Sonnet and DeepSeek v3 led early testing across 500,000 comparisons.
Adobe just rolled out Acrobat Studio, combining Acrobat, Adobe Express, and agentic AI into a single hub. Its new PDF Spaces feature turns file collections into interactive knowledge bases where you can assign custom roles like "analyst" to assistants, then immediately turn insights into slides or infographics. Early access starts at $24.99 a month.
Microsoft AI chief Mustafa Suleyman just published a warning about the rise of Seemingly Conscious AI, arguing systems that mimic consciousness will arrive within two to three years using today's tech. His main concern is that convincing illusions will trigger severe psychological attachment and misplaced calls for AI rights. Why it matters: Suleyman wants the industry to draw a hard line around giving AI personality without personhood.
OpenAI just launched ChatGPT Go in India, a new subscription tier priced at ₹399 ($4.60) per month with support for UPI payments. The plan offers 10x the message, image generation, and file upload limits of the free tier, alongside improved memory retention. It is currently exclusive to India, with plans to expand to more countries later.
OpenAI CEO Sam Altman just admitted that investors are overexcited and the industry is currently in an AI bubble, though he still calls it the most important technology in decades. He revealed ChatGPT now hits 700 million weekly users with servers running at full capacity, while explicitly shutting down any plans to turn the bot into an anime companion.
NVIDIA just dropped the first major update to Project G-Assist, its experimental on-device assistant for PC gamers. A new lightweight model cuts VRAM usage by 40%, bringing local voice and text controls to any RTX card with at least 6GB VRAM. It also integrates a mod.io plug-in hub you can browse via chat.
Qwen just introduced Qwen-Image-Edit, a 20B-parameter foundation model built specifically for advanced image and text editing. It can directly modify Chinese and English text inside images while preserving fonts, alongside pulling off full 180-degree novel view synthesis and localized element tweaks. You can already try it inside Qwen Chat.
ElevenLabs just rolled out the Eleven Music API, pitched as the first music API for developers trained on licensed data and fully cleared for commercial use. It lets teams plug AI music directly into their products and workflows. The company also noted that creators have already generated over 750k songs since Eleven Music debuted.
Google Labs just doubled the monthly AI credits for Google AI Ultra subscribers from 12,500 to 25,000, giving creators twice the runway for its filmmaking tool Flow and image experiment Whisk. Flow has already generated over 100 million videos since May, while Whisk is expanding into 77 more countries. Why it matters: Google is sweetening its top-tier AI subscription with serious creative firepower to keep up with competing video generators.
Grammarly just rolled out nine specialized AI agents in its docs writing surface, including a grader bot that predicts paper scores based on course info and instructor data. Other agents handle citation finding, tone adjusting, and plagiarism detection. The plagiarism and AI detector agents are currently limited to Pro users.
Atlas AI just partnered with Google Cloud to launch an AI-native game creation platform running on Vertex AI. Developers can generate clean, game-ready 3D assets and environments directly from mood boards and text prompts. It is designed for production pipelines rather than one-off assets and is currently in closed beta for game studios.
Microsoft just added a native =COPILOT function to Excel for Windows and Mac, bringing LLMs directly into the formula bar. You can write natural language prompts and reference cell ranges to classify feedback, generate summaries, or spit out multi-column tables that update automatically when source data changes. Note that Microsoft later announced the formula is slated to retire on September 14, 2026, steering users back to the side pane.
Nvidia just dropped Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer model slimmed down to fit on a single A10 GPU. Developers can toggle reasoning traces on or off using simple /think tokens and dial in exact thinking budgets. It delivers up to 6x faster inference than similar transformers while outperforming Qwen3-8B across key benchmarks.
Researchers from UC Berkeley, Stanford, and Databricks just introduced GEPA, a prompt optimization method that ditches reinforcement learning. Instead of thousands of trial-and-error runs guided by simple scores, GEPA has models reflect on detailed execution logs and error traces, beating traditional methods with up to 35 times fewer trials.
Google just added an audio listening feature to Google Docs powered by Gemini, letting users hear their documents read aloud with natural-sounding voices and adjustable speeds. You can find "Listen to this tab" under Tools > Audio, though rollout is currently limited to Workspace business, enterprise, and Gemini add-on tiers.
Former Twitter CEO Parag Agrawal just unveiled Parallel Web Systems with $30M in backing from Khosla Ventures and others to build an "Internet for AI." The startup launched its Deep Research API, aiming to replace human-centric browsing with a programmatic web tailored for autonomous AI agents.
Meta is reportedly preparing to reveal its new Hypernova smart glasses in September, lowering the starting price from $1,000 to $800. The device features a screen on one lens for apps and alerts, paired with wrist controls and a voice AI query tool. It marks Meta's clearest attempt yet at creating hardware that could eventually challenge the smartphone.
OpenAI just tweaked GPT-5's personality to make conversations feel warmer and more approachable after feedback that it felt too stiff. The update introduces subtle touches like "Good question" without increasing sycophancy, according to internal testing. The rollout takes up to a day to complete.
Google just introduced Flight Deals, a conversational AI feature inside Google Flights. Instead of constantly toggling dates and filters, travelers can describe their dream trip in plain English, and the tool finds matching bargain flights across airlines. It rolls out in beta over the next week.
Higgsfield just introduced Product-to-Video within its Draw-to-Video platform, allowing creators to insert objects directly into scenes without relying solely on complex prompts. Users simply drop an item into a shot and sketch quick cues, while the AI renders a cinematic sequence. It could drastically lower costs for branded content and commercial shoots.
ElevenLabs just rolled out a new Video-to-Music flow inside ElevenLabs Studio. Powered by its Eleven Music model, it generates custom soundtracks tailored to your video in a single click, and lets you layer in voiceovers and SFX on the spot. It turns video audio production into a one-stop workflow.
Sam Altman just addressed the backlash over GPT-5's recent launch, admitting OpenAI "totally screwed up" aspects of the rollout and bringing back 4o for paid subscribers. He revealed ChatGPT now reaches 700 million weekly users, while warning that investors are overexcited and the industry is currently in an AI bubble.
NVIDIA just released Granary, an open-source multilingual speech dataset spanning around a million hours of audio across 25 European languages. Alongside it, the company launched Canary-1b-v2 for high-accuracy translation and Parakeet-tdt-0.6b-v3 for fast, real-time transcription on Hugging Face.
Anthropic just overhauled its Usage Policy to keep up with agentic tools like Claude Code, with changes taking effect September 15. The update sets clear lines against malware and infrastructure attacks, while replacing a blanket political ban with targeted rules that allow policy research but ban voter targeting.
Anthropic just gave Claude Opus 4 and 4.1 the ability to actively shut down conversations in extreme cases of persistent abuse or harmful requests. The feature stems from the company's exploratory research into AI welfare, giving models an exit when redirection fails. Users hit with a cutoff can't message that thread anymore, though they can still start new chats.
