Google just admitted in an antitrust court filing that the "open web is already in rapid decline," arguing that breaking up its ad business would accelerate the crash for publishers. That statement starkly contradicts its public claims that web traffic is thriving under AI search. Google later pushed back, arguing the line was cherry-picked and only applied to display ads.
Anthropic just endorsed California’s SB 53, a bill targeting frontier AI models trained with at least 10^26 FLOPS. Instead of rigid technical mandates, the bill requires big labs to publish safety frameworks, report critical incidents within 15 days, and protect whistleblowers, all while exempting startups. Anthropic says mandatory transparency keeps labs from cutting safety corners to compete.
Anthropic just agreed to pay at least $1.5 billion to settle a class-action lawsuit accusing it of training models on pirated books. The deal pays out roughly $3,000 per work across an estimated 500,000 titles and requires Anthropic to destroy all downloaded files. Why it matters: It marks the largest copyright payout in U.S. history, setting a costly precedent for AI training data.
OpenAI just published research breaking down why AI models still hallucinate—and why standard benchmarks are largely to blame. Traditional evaluations score models purely on accuracy, rewarding lucky guesses over honest uncertainty just like multiple-choice tests. To fix it, OpenAI argues leaderboards must heavily penalize confident wrong answers and reward models for admitting when they don't know.
Eliza Labs just filed an antitrust lawsuit against X, claiming the platform coaxed technical details out of them for their ElizaOS agent framework, abruptly banned their accounts, and copied the tech for xAI. Eliza is seeking treble and punitive damages, though legal experts note that antitrust claims around open-source tools face an uphill battle.
Anthropic just updated its consumer privacy policy to train AI models on user chats and coding sessions, retaining data for up to five years. The change takes effect September 28 across Free, Pro, and Max tiers with opt-in enabled by default, though enterprise and API accounts remain exempt.
Anthropic just updated its consumer terms and privacy policy, giving Free, Pro, and Max users a choice to let their chats and Claude Code sessions train future models. Opting in extends data retention from 30 days to five years, while commercial and API customers remain untouched. Users have until October 8, 2025, to decide in their privacy settings.
OpenAI and Anthropic just shared results from a first-of-its-kind joint safety evaluation, running their internal alignment tests on each other's flagship models. Claude 4 excelled at resisting system prompt extraction but refused up to 70% of hallucination probes, while OpenAI o3 held the line best against jailbreaks. Having fierce rivals probe each other's vulnerabilities is a rare, refreshing win for frontier model transparency.
Anthropic just released a threat report exposing how bad actors are weaponizing its AI. In one wild case, a hacker used Claude Code to automate network infiltration, extract data, and calculate ransoms across 17 organizations. The company banned the accounts and deployed new classifiers, but it shows agentic AI is officially in criminals' hands.
Anthropic just formed a bipartisan National Security and Public Sector Advisory Council packed with heavyweights, including former Acting Defense Secretary Patrick Shanahan and former CIA Deputy Director David Cohen. The group will guide the AI lab's defense and intelligence rollouts in areas like cyber defense and analysis. It is a clear push to cement frontier AI inside the U.S. government.
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.
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.
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.
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.
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.
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.
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.
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.
