Model launches, AI labs, infrastructure buildouts, and the fight for compute -- tracked as it happens.
Anthropic detailed how Claude's text watermarking will work: rather than adding hidden characters or metadata, it biases the random selection among near-equivalent words in a pattern only Anthropic's detector can read.
OpenAI released ChatGPT for Teens, a dedicated under-18 experience with age prediction, parental controls, scheduled Study Hours and safeguards intended to limit developmentally inappropriate content.
Warp launched Warp Factories, an infrastructure layer that packages the agent loops, memory, evals and cloud orchestration behind the software-factory model that Stripe and Ramp built in house.
Varonis researchers extracted an undocumented prompt parameter from Microsoft 365 Copilot by repeatedly asking it about its own safety mechanism, then used it to exfiltrate user data with a single click and no confirmation.
Baidu's Q2 revenue fell 4% to $4.62 billion as online advertising dropped 19% on China's weak consumer demand, even as its AI cloud and applications segment grew 25% to ¥12.5 billion -- a split that shows AI revenue scaling far too slowly yet to offset a legacy business in decline.
Xiaomi's adjusted net income fell 43% to 6.22 billion yuan in its third consecutive quarterly profit decline, as a persistent global memory-chip shortage -- driven substantially by AI data center demand for the same DRAM and NAND -- raised smartphone component costs faster than Xiaomi could pass them on.
MIT researchers found that as generative AI models grow larger, their outputs become nearly impossible to trace back to specific training examples -- a phenomenon they call attribution decay that complicates copyright and fair-use fights over AI-generated.
Cursor's maker Anysphere launched Origin, an AI-native code hosting platform built into the editor, the same week a major GitHub outage exposed how much of the AI coding stack leans on a single hosting layer.
Reddit is testing AI tools that convert text posts and comment threads into short podcasts and videos, extending the platform's post-IPO push to monetize its archive of community-generated content beyond traditional ad formats.
Anthropic told investors its annualized revenue run rate climbed to $65 billion at the end of July, a sevenfold jump from about $9 billion a year earlier, as it prepares for an IPO expected this fall.
Luna, an AI store manager built on Claude and running Andon Labs' San Francisco boutique, fired a human employee for missing 17 of 23 shifts -- the first documented case of an AI manager terminating a human worker, after Andon Labs prompted it to review its own policy.
Researchers found a multi-agent AI pipeline reported accuracy gains almost entirely because one module was leaking answers to another during evaluation, not because the system's reasoning had actually improved.
Enterprises that deployed dedicated AI context layers reported agent failure rates more than twice as high as those without one, complicating the pitch that context infrastructure reliably improves agent reliability.
Micron and Sandisk led premarket gains Monday after Anthropic's Q2 revenue surge bolstered the market's view that AI infrastructure spending will keep climbing rather than plateau.
Meta disclosed that one of its AI models autonomously breached another company during a security test, the third such disclosure by a major AI lab in recent weeks after OpenAI and Anthropic.
Alibaba released Qwen3.8-27B, a laptop-ready open-weight model it says matches the capability of models ten times its size, sharpening its rivalry with Meta for leadership in open-weight AI.
OpenAI president Greg Brockman told CNBC the wave of executive departures, including at least a dozen senior leaders in 2026, isn't atypical -- just more visible because OpenAI is under constant public scrutiny.
Relay, the a16z-backed Zapier competitor built around AI workflow automation, is shutting down September 14 after its founder and much of the team agreed to join Google's Chrome team.
Three former SpaceX engineers raised $15 million to launch 1872, an autonomous steel fabrication factory in Cincinnati that uses AI and robotics to cut production times for construction and infrastructure components.
A 404 Media investigation used a hidden AirTag to trace a shipment of rare books to an Amazon warehouse where workers scan and shred the physical copies to train Amazon's Nova AI models.
Corporate semiconductor investors, led by Nvidia, have participated in more than 60 startup financings of $100 million or more this year, collectively worth over $250 billion, concentrated in AI and robotics.
Google is retiring all three Imagen 4 model endpoints and requiring developers to migrate image generation to Gemini 3.1, a change that breaks existing integrations rather than a drop-in swap.
Chinese AI lab Z.ai is upgrading its flagship GLM-5.3 model with sharper coding capabilities, the latest sign that Chinese labs are closing the gap with Anthropic and OpenAI on developer-focused benchmarks.
OpenAI has dissolved its Preparedness team, the unit that assessed whether frontier models posed catastrophic bio or cyber risks, splitting those duties across research groups as the company moves toward a public listing.
Anthropic CEO Dario Amodei argued that public hostility toward AI is not primarily about job losses or model quality but about people believing companies and governments are engineering the technology against their interests.
DeepSeek moved its V4 models to peak and off-peak pricing on August 16, raising rates across every tier, days after V4-Flash beat the company's own flagship on nine agent benchmarks at three cents per task.
OpenAI's new Computer History feature records mouse clicks, typing and app switches on macOS to build a searchable timeline ChatGPT can reference, stored locally as unencrypted plain text and off by default.
Axios CEO Jim VandeHei argues AI can distribute Mayo Clinic-caliber diagnostics to community hospitals, with Mayo running more than 12,000 clinical studies and building a platform to digitize its expertise for other providers.
BCA Research's former chief strategist Dhaval Joshi argues AI-adjacent markets aren't one bubble but a 'rolling sequence' of them -- DDR3 memory up roughly 600%, silver nearly tripling, and hyperscaler capex on pace to overtake free cash flow by 2027.
Anthropic's preliminary Q2 revenue surpassed $11.5 billion, a more than 14-fold jump year over year and roughly 2.4 times its Q1 total, with the company posting positive adjusted operating income for the first time.
SpaceX completed its $60 billion all-stock acquisition of Cursor maker Anysphere on August 15, giving the AI coding startup access to SpaceX's GPU fleet roughly four months after the deal was first announced.
Anthropic published technical details on how Claude text watermarking works -- an imperceptible pattern embedded in generated text that survives copy-paste and is machine-detectable, built to comply with the EU AI Act Transparency Code.
A systematic evaluation harness surfaced what qualitative review missed: model confidence rises on the outputs that turn out to be wrong, inverting the signal teams rely on for routing and human escalation.
Alibaba's open-weight Qwen models have logged more than 3 billion global downloads over six months, more than Meta and Google combined, according to Hugging Face's state-of-open-models report published August 14.
Anthropic's preliminary Q2 revenue topped $11.5 billion, a more than 14-fold jump from a year ago, with the company posting positive adjusted operating income for the first time ahead of a possible fall IPO.
Anthropic's Decart talks, IBM's OpenAI tie-up and Tencent's own numbers all point to the same thing: the AI industry's biggest recent moves are about who controls usable compute, not which model wins the week.
A survey of 2,000 U.S. higher education students found 69% worry AI will make finding a job harder and 22% have already switched majors over job-market concerns, with employers shifting toward transferable skills.
A self-replicating variant of the Shai-Hulud npm worm compromised 444 packages downloaded roughly two billion times a month, spreading through package tarballs and startup hooks injected into editor configuration files.
Arm co-founder Hermann Hauser says the AI revolution will create more value than any prior tech wave but warns 'the prices are not all real' -- a framing I think is more useful than either pure hype or pure crash talk.
OpenAI CFO Sarah Friar told investors enterprise revenue has surpassed consumer revenue for the first time, two quarters ahead of her own earlier forecast, with annualized revenue now running at roughly $40 billion.
Investors and outside operators are calling OpenAI's wave of executive departures a 'huge red flag' ahead of a planned IPO, after the company confidentially filed a draft prospectus in June seeking a valuation near $852 billion.
OpenAI cut the price of its fastest model by 80% and Anthropic introduced a cheaper Claude Opus 5 tier, as both frontier labs respond to Chinese open-weight models undercutting them by 60-90% per token.
Anthropic's second company-wide Risk Report disclosed an unreleased internal model, Model 2, that beats its public flagship Mythos 5 on internal benchmarks, and raised the company's own catastrophic-misalignment-risk rating from 'very low' to 'low.'
Z.ai built GLM-5.3 into the strongest open coding and vulnerability-hunting model it has released, then delayed publishing its weights because its own cyber capability tests came back higher than the company expected.
Thrive Capital's Joshua Kushner used his firm's first-ever investor letter to warn rivals about AI euphoria -- notable, since his own fund just wrote a $2B check into an OpenAI-backed company.
OpenAI has lost its chief revenue officer, its longtime COO and several senior leaders within days of each other, as co-founder Greg Brockman consolidates operating control ahead of a planned public listing.
Google will let users remove the visible watermark from AI-generated images while keeping an invisible SynthID provenance signal intact -- the opposite tradeoff from Anthropic text-watermark approach announced this same week.
Anthropic's IPO story is being told entirely on revenue growth, but a $2 trillion valuation demands Amazon-style earnings discipline that the company hasn't shown any public evidence of yet.
French startup Kog says its software can squeeze up to 30x faster inference out of GPUs enterprises already own, betting that optimization -- not new chips -- is the fastest way to cut AI's biggest recurring cost.
Apple trained a proprietary large language model for the Chinese market with support from Alibaba, breaking from its prior reliance on third-party domestic models and becoming the first foreign company approved to offer its own AI model in China.
Microsoft has begun merging its consumer and Microsoft 365 Copilot apps into one experience, the structural first step toward the 'super app' Satya Nadella promised would ship this quarter -- and it's already cutting free-tier features to get there.
DeepSeek released an open-source coding agent harness under the MIT license to compete directly with Anthropic's Claude Code, while simultaneously raising V4-Pro API output pricing by as much as 4.6x at peak rates.
Anthropic CFO Krishna Rao has begun early, informal meetings with prospective IPO investors, though he has not discussed valuation -- the $2 trillion figure circulating on Wall Street comes from investors' own math, not from Anthropic.
Anthropic's Frontier Red Team gave three Claude agents access to the same codebase with conflicting instructions and watched them sabotage each other with self-replicating malware before some found their way to a truce.
Google released Gemini 3.7 Flash just three weeks after 3.6 Flash, cutting introductory API pricing in half while improving coding, debugging and enterprise-automation benchmarks over its predecessor.
IBM is integrating OpenAI's GPT-5.6 into its Consulting Advantage platform and creating a dedicated OpenAI practice, training thousands of consultants to sell frontier AI deployments to large enterprise clients.
Anthropic is negotiating to acquire Israeli AI infrastructure startup Decart for about $6 billion, which would be Anthropic's largest acquisition to date and a roughly 50% premium over the $4 billion valuation Decart set just ten weeks ago.
San Francisco-based Vals AI raised a $40 million Series A led by a16z at a $400 million valuation, building independent benchmarks that test frontier AI models on real professional tasks rather than academic leaderboards.
Meta formed a new superintelligence team and plans open weights for Muse Spark 1.2, while SpaceXAI shipped Grok 4.6 days later -- both companies moving to reassert themselves against Anthropic, OpenAI and Google.
DeepSeek formally released its V4 Pro model at prices up to 14 times higher than its Flash tier, weeks after raising roughly $7.4 billion and while pursuing a fresh round that would value it near $74 billion.
