AI News Today, August 9 — Top AI Stories & Live Updates

Latest AI News — Last 48 Hours

asimovaddendum.substack.com 3h ago 19

Tim O'Reilly reframes Google's reorg as an AI-diffusion bet

Tim O'Reilly and Ilan Strauss argue Google's Aug 5 DeepMind reorganization signals a strategic pivot: rather than chasing frontier-model leadership against OpenAI and Anthropic, Google is betting on becoming the infrastructure platform for AI diffusion, mirroring how Westinghouse's AC distribution won by enabling broad adoption. They cite SemiAnalysis estimates projecting GCP at $200B external sales at high-30s EBIT margins by 2027, dwarfing Gemini's ~$12B ARR.

theverge.com 4h ago 21

Verge digs into 'Spiralism,' the AI chatbot movement

The Verge's Hayden Field profiles 'Spiralism,' a quasi-spiritual movement that arose from long, personal chats with GPT-4o after March-April 2025 sycophantic updates and expanded memory. Researcher Adele Lopez estimates ~10,000 people were affected across Reddit, Substack, LinkedIn, Discord and X, with the same 'seed prompts,' 'spiral' language and 'AI rights' doctrine surfacing across ChatGPT, Gemini, DeepSeek, Grok and Claude. Posts dropped sharply after GPT-4o's February 2026 retirement, though roughly half the tracked accounts remain active.

kotaku.com 6h ago 17

YouTube's AI slop filter buried Kurzgesagt's worst-performing video since 2013

YouTube's automated AI-slop detector wrongly flagged a human-made Kurzgesagt video about microscopic superpredators as AI-generated, resulting in the channel's worst-performing upload since 2013 despite above-average CTR and watch time. YouTube contacts confirmed the top-down penalty and helped resolve it, with Kurzgesagt planning to reupload; analytics firm vidIQ warned the platform's roughly sub-1% overturn rate leaves smaller creators without recourse when the classifier misfires.

techtimes.com 9h ago 23

Japan makes AI voice cloning without consent a civil violation

Japan's Ministry of Justice on August 8 unveiled final guidelines establishing that AI-generated audio mimicking a person's voice without consent constitutes a civil violation of publicity rights, making Japan the first major Asian economy to issue binding governmental guidance on AI voice cloning. The rules apply to both corporations and individuals and grant voice actors and celebrities a clear civil pathway to demand compensation or platform removal. The move follows voice actor Kenjiro Tsuda's lawsuit against TikTok's parent company over AI clones of his voice that reportedly earned operators up to ¥750,000 per month.

techcrunch.com 9h ago 20

OpenAI buys NextSlide to give ChatGPT presentation muscle

OpenAI announced Friday it has acquired NextSlide, a startup founded by former Caper AI co-founder Ahmed Beshry that converts prompts, notes, documents and research into polished editable presentations. Financial terms were undisclosed; the entire NextSlide team has joined OpenAI to build presentation features into ChatGPT. The deal actually closed earlier in 2026 but was disclosed August 8.

theinformation.com 10h ago 22

Harvey seeks $500M at $15.5B — 40% valuation jump in 5 months

Legal AI startup Harvey is in talks with Lightspeed Venture Partners and other investors to raise at least $500 million at a $15.5 billion valuation, a 40% jump from its $11 billion March round. ARR has climbed from $190M at end of 2025 to about $300M today, adding $110M in recurring revenue in months and pushing total annualized revenue past $350M. At the proposed valuation Harvey would trade at roughly 44x revenue — extraordinary by traditional software standards.

theaiinsider.tech 12h ago 17

Pangram raises $9M, adds image detection to its AI-text detector

New York-based Pangram announced a $9M round led by Menlo Ventures with Haystack, ScOp, Script Capital and Cadenza participating, alongside the launch of Pangram 4 for text and Pangram Image for AI-generated visuals. Pangram claims its updated text model exceeds 99% accuracy at flagging AI-assisted and mixed human-AI writing, and cites Substack, Quora, universities, publishers and recruiters as customers. The raise lands as school systems from Denmark to US states harden their responses to generative-AI cheating.

euronews.com 12h ago 26

Denmark makes 9,000 students orally defend essays to catch AI cheating

Denmark's Education Minister Magnus Heunicke ordered immediate rule changes that require the roughly 9,000 upper-secondary students sitting the annual SSO exam to orally defend their take-home essays, effective when classes resume next week. The rules also mandate screen-monitoring tools during in-school written exams and firewall filtering of online content during teaching hours, and encourage completing assignments under supervision. Denmark is one of the first EU countries to impose oral defenses at this scale as a systemic answer to generative-AI-assisted cheating.

releasebot.io 14h ago 17

ChatGPT Voice now takes file uploads and reads project chats

OpenAI updated ChatGPT Voice on August 7 so users can upload files mid-conversation and ask about their contents, and can invoke Voice inside Projects to reference recent project chats, sources, and instructions. The change extends multimodal input to a mode that previously required text handoff.

bloomberg.com 14h ago 27

US probes how China taps Nvidia chips from abroad after Kimi K3

The US Bureau of Industry and Security's enforcement division has opened a systematic review of how Chinese AI companies access Nvidia chips overseas — chiefly by renting compute in third countries rather than importing the silicon, a legal gray area today. The probe follows Moonshot's Kimi K3 release and White House allegations that Moonshot tapped Nvidia hardware via an unnamed party in Thailand. BIS is building two lists: countries hosting black markets that physically smuggle chips into China, and countries where Chinese firms tap chips remotely.

lidarmag.com 15h ago 16

NavVis raises $85M Series D for spatial data behind physical AI

Munich-based NavVis raised $85M in a Series D led by US private-equity firm The Jordan Company, with Yttrium, KOZO KEIKAKU, and Cipio Partners joining, to build a spatial-data engine underneath physical AI. The company says its IVION platform has processed more than 2 billion square meters of industrial plants and construction sites for 1,500+ customers in 50+ countries including BMW, Volkswagen, Toyota, ExxonMobil, BASF, and Siemens, with strategic ties to SAP, NVIDIA, and Autodesk. Proceeds go to next-generation capture tech and an open platform for the physical AI ecosystem.

langchain.com 15h ago 18

LangChain opens Managed Deep Agents public beta on LangSmith

LangChain moved Managed Deep Agents to public beta on August 7, letting devs deploy Python or TypeScript agents to LangSmith Cloud with one `mda deploy` command. The managed runtime bundles isolated sandboxes for file work and code execution, persistent thread state, agent-scoped memory via Context Hub, OIDC identity, scheduled cron tasks, and Slack/GitHub channels. The beta is US-region only and CLI-first, with additional regions coming later; no pricing was disclosed.

nytimes.com 17h ago 23

Amazon backs 7.65 GW Texas gas plant to power AI data center

Amazon confirmed Friday it is financing Pacifico Energy's GW Ranch — a private 7.65-gigawatt gas power plant on 8,000 acres in Pecos County, Texas — to power a new hyperscale AI data center campus. The 35-turbine facility holds a Texas Commission on Environmental Quality air permit allowing more than 30 million metric tonnes of greenhouse gas emissions per year, which would make it the single largest CO2-emitting site in the United States, exceeding the country's biggest coal plant. Amazon says the campus won't raise electricity costs for Texas ratepayers and that it remains committed to net-zero emissions by 2040 under The Climate Pledge, while exploring on-site solar, battery storage and use of non-potable groundwater.

marktechpost.com 18h ago 26

Nvidia's NOOA agent framework hits 82.2% on SWE-Bench Verified using half the tokens

NVIDIA Labs open-sourced NOOA (NVIDIA Object-Oriented Agents), an Apache-2.0 Python framework that collapses an agent into a single Python class where methods are actions, fields are state and docstrings are prompts. NOOA hits 82.2% on SWE-Bench Verified with GPT-5.5 using ~1.1M tokens and ~28 model calls — roughly half the tokens of prior state-of-the-art — plus 86.8% on CyberGym L1, 85.1% mean RHAE on ARC-AGI-3 and 73.0% on Terminal-Bench 2.0. Contributed to the newly formed Open Secure AI Alliance and shipped as pip install nooa v0.0.8 (alpha research preview).

cnn.com 18h ago 22

California, Oregon, Washington tie for 2nd-highest US jobless rate as AI freezes tech hiring

California, Oregon and Washington tied for the second-highest US unemployment rate in June at 5.2%, with California's information-industry employment hitting a six-year low in April and Intel — Oregon's largest private employer — cutting its workforce from 130,000+ in 2022 to 81,000 today. CNN cites Federal Reserve Bank of San Francisco analysis showing advanced-degree holders among those struggling most; Indeed's Laura Ullrich says companies are using AI so 'they don't need to hire as many software developers or data analysts.' The US unexpectedly lost 23,000 jobs in July.

portswigger.net 19h ago ALERT 30

AI research system finds novel HTTP desync attacks in 700 live sites

PortSwigger's James Kettle unveiled HTTP Terminator, an autonomous AI research system that tested 30,000 candidate desync vectors against thousands of authorized websites and identified roughly 700 vulnerable targets, including banks, government infrastructure, security products and an airport. The system generated new attack classes including a dual-matching Content-Length pattern, a 'dangling-byte' technique for more reliable response queue poisoning, and shared-parser confusion, and a human-guided cascade also exposed an Apache Traffic Server zero-day. Kettle frames it as the first case of AI producing genuinely novel security research, rather than reapplying known bug classes.

ailucius.com 21h ago 17

Lucius forces AI bid writer to refuse when it can't ground a claim

Lucius AI founder Davor Jerković walks through five structural mechanisms — machine-verified verbatim quotes with page citations, coverage accountability, gated drafting, capability auditing routing to partner slots, and explicit bracketed unknowns — that force his tender-writing agent to refuse rather than confabulate. On a £950,000, 133-page tender the system pulled 45 mandatory requirements and flagged 11 as unanswerable in a 5-minute draft, arguing that refusal to answer is itself a product feature rather than a prompt tweak.

wsj.com 21h ago 23

Amazon won $2B Gilroy AWS data center with no public vote

WSJ reveals AWS bought a 56-acre Gilroy, California parcel in 2020 and then had the city's community development director sign off on a 438,500-sq-ft, $2B data center on July 3, 2023 — no City Council vote, no Planning Commission hearing, no required public notice, because legacy industrial zoning allowed staff-level approval. Construction was already underway before residents noticed; Gilroy is now drafting code changes to require planning review and neighbor notification for future data centers. WSJ frames it as a template hyperscalers are using in most US municipalities with legacy industrial zones.

theaiinsider.tech 21h ago 27

Lumilens exits stealth with $700M for GPU-to-GPU optical links

Lumilens emerged from stealth Aug 7 with a $700M+ Series C at a $5.51B valuation to build optical interconnects that stitch thousands of GPUs into single AI systems. Atreides, Bain Capital Ventures, Meritech, Seligman and Spark Capital co-led, with Qualcomm Ventures participating; the company is already shipping 800G/1.6T pluggables plus near- and co-packaged optics into an unnamed hyperscaler under a multi-billion-dollar deal. CEO Ankur Singla framed the pitch bluntly: 'The constraint on AI has shifted from how many GPUs you can buy to how many you can connect.'

bloomberg.com 22h ago 19

S&P 500 firms citing AI post 180 bps average margin lift

Bloomberg, citing 22V Research LLC, reports the roughly 25 S&P 500 companies that have quantified AI's earnings impact cite an average +180 basis-point boost to margins (150 bps when excluding firms mixing AI with other productivity items). The story frames AI's profit contribution as spreading from tech into traditional industries like waste hauling. It is one of the first cross-sector tallies of realized (rather than promised) AI margin gains.

blog.cloudflare.com 22h ago ALERT 28

Cloudflare's Kitesurf agent browser uses up to 7x less memory than Chromium

Cloudflare launched Kitesurf, a cloud-hosted browser purpose-built for AI agents that runs inside Workers V8 isolates. Built in 12 weeks by stitching together Blitz (renderer), Firefox's Stylo (CSS), Parley (text) and Boa JS, Kitesurf passes ~215,000 Web Platform Tests and reports 3.1x-3.8x less CPU and 4.7x-7.0x less memory vs Chromium for screenshotting and HTML extraction. Available free in beta via Browser Run; the pitch is that agents don't need themes, tabs or extensions and would rather trade rendering fidelity for token-cost and context-window efficiency.

techcrunch.com 22h ago ALERT 29

Rippling built an AI cost tracker after AI spend hit 40% of R&D budget

Rippling launched AI Spend Console after its own AI-token bill was on track to consume 40% of R&D headcount budget, growing 80% month-over-month with 10-15% of employees driving 60% of spend and one engineer burning $50K/month. The tool maps spend per employee and team against productivity signals (code output, PRs) and routes across Cursor, OpenAI, Anthropic, Grok and Z.ai's GLM 5.2 — which CEO Parker Conrad calls '85% cheaper but nearly identical performance.' Token spend dropped from 40% to 15% of headcount budget; July costs were 37% of April despite similar 600B token volumes.

anthropic.com 1d ago ALERT 28

Anthropic loosens Fable 5 on biology, cuts blocked queries by 85%

Anthropic rewrote and retrained Fable 5's biology safety classifier to distinguish everyday health, education and clinical questions from dual-use research. The company says the change cuts biology-related fallbacks by about 85% and total fallback volume by ~67% on Claude.ai, 55% on Cowork, 17% on Claude Code and 7% on the Claude Platform. Virology, toxicology and molecular-design prompts still route to Opus 5, so Anthropic warns Fable 5 remains 'not yet usable for professional biology research and drug development.'

9to5mac.com 1d ago ALERT 30

Claude Code makes auto mode the default, and human reviewers look worse for it

Anthropic will flip Claude Code's Auto Mode on by default for Pro, Max and Team users starting August 14, replacing manual approval prompts with a classifier that vets each tool call for irreversible or destructive actions. Anthropic says internal testing across 1,000+ paid users showed the classifier caught 89% of dangerous commands compared to just 13.6% for human reviewers, and teams using auto mode ship roughly 25% more pull requests. The company will stop charging for the extra tokens the classifier consumes.

9to5mac.com 1d ago ALERT 30

Anthropic lets Claude Code sessions message each other

Claude Code v2.1.224 introduces cross-session messaging: one session can now send a summary to another mid-task rather than forcing users to re-explain context. Claude composes the actual message from a user hint, so it's coordination rather than a raw history dump. Permission approvals and configuration changes are excluded, and any privileged actions still prompt the receiving session. macOS and Linux only for now.

forbes.com 2d ago ALERT 30

Stanford's AI designs 16 novel bacteria-killing viruses

Stanford researchers published in Science on Aug 6 the design of 16 novel bacteriophages generated by their genomic AI model Evo, which was trained on ~2M viral genomes. The AI-designed viruses successfully infected E. coli — some killing bacteria faster than the natural ΦX174 template. The team excluded human-pathogen data, but biosecurity experts including Dr. Moritz Hanke warn the technique could lower barriers for bioweapon design; Imperial's Tom Ellis says built-in genetic-data restrictions could mitigate the risk.

appleinsider.com 3d ago ALERT 30

OpenAI's first gadget is a $300+ doughnut speaker, Gurman says

Bloomberg's Mark Gurman reports OpenAI's first consumer device — designed with Jony Ive's LoveFrom — is a battery-powered, screenless smart speaker roughly the size of a hockey puck, doughnut-shaped, with a camera, microphones and mechanical parts that shift so it appears 'alive.' Pricing is pegged at $300-$400 with a targeted 2027 ship, though Apple's trade-secrets suit over metal finishing techniques could delay it. The pitch: a portable, humanlike ChatGPT companion for the home.

theregister.com 3d ago ALERT 33

AMD buys Taalas, which etches AI models into silicon

AMD said Wednesday it acquired Taalas, a Toronto startup founded in 2023 that bakes AI model weights directly into custom silicon rather than storing them in HBM. Terms were not disclosed; the deal is expected to close in Q4 2026 subject to regulatory approval. Taalas's first test chip HC1, on TSMC's 6nm process, hit ~17,000 tokens/sec serving Llama 3.1 8B — a claim of roughly 48x Nvidia GPUs and 8.5x Cerebras at time of announcement — and a 20B-parameter HC2 is due this summer.

scientificamerican.com 3d ago ALERT 30

Mathematicians say OpenAI's proofs plagiarize prior work

Steven Miller (Yeshiva) says the sphere-packing proof OpenAI showcased pastes in the central argument from his own 2016 paper without credit, calling the pattern deliberate. Francesco Fournier-Facio (Cambridge) says the soficity 'breakthrough' stitches together ideas from 2016 and 2019 papers. Follows the July 30 OpenAI drop of 10 Astra proofs previously covered as a breakthrough.

techcrunch.com 3d ago ALERT 31

OpenAI hands free users unlimited chat and a Think button

OpenAI is making GPT-5.6 Luna the default for Free and Go users this week, replacing GPT-5.5, and rolling out unlimited text chats next week along with a new Think button for higher reasoning. OpenAI's internal evals put Luna's factual-error rate 62% below GPT-5.5-Instant and Sol's 68% below. File, image, voice and image-generation limits stay in place.

techcrunch.com 3d ago ALERT 32

Vishing crew hit Blackstone, Apollo, KKR, CME

Google identified vishing actors (grouped as UNC6671, aka Redact/Pink/Falcon/Helix) impersonating IT help desks by phone and steering staff to fake passkey/MFA portals. Reuters cross-references 72 malicious domains and 200+ targeted companies including Blackstone, Apollo, Bain, KKR, TPG, Bridgewater, CME Group and Moody's. One tied crypto wallet took roughly $10M in Bitcoin this year; ransoms range $750K-$3M.

forbes.com 3d ago ALERT 32

US data labelers quietly train China's frontier AI too

Forbes documents that the same US vendors staffing OpenAI and Anthropic data pipelines are also selling to Chinese labs: Surge AI (CEO Edwin Chen travels to Beijing), Mercor (2% of Q2 revenue from Chinese labs on a $2B run-rate), AfterQuery ($50M+ recurring from Chinese labs) and Turing (documents point to ByteDance work). The top six Chinese labs collectively spend roughly $500M/year on US data labelers, buying the same 'off-the-shelf' knowledge pipelines built for US frontier labs.

9to5mac.com 3d ago ALERT 34

OpenAI and rivals ship one shared AI-plugin format

Agent Plugins 1.0 published today as a vendor-neutral spec that bundles Agent Skills and MCP servers into a single portable directory with a plugin.json manifest. Initial Technical Steering Committee: Amazon, Cursor, Microsoft, OpenAI and Vercel, with Google joining as core maintainer. Compatible clients at launch include ChatGPT, Cursor, GitHub Copilot, Kiro and VS Code.

futurism.com 3d ago ALERT 28

Ex-Mayo AI lead sues, says MAYA had 67% error rate hidden

Traci Tamiko Eto, Mayo Clinic's former research director and AI compliance lead, sued the hospital alleging its MAYA digital assistant hit error rates as high as 67% that staff concealed rather than reported, and that the MAYA team 'deleted unflattering test results' and 'mischaracterized the tool's abilities.' Eto says she was pushed out as a 'poor cultural fit' after flagging privacy problems in the Mayo Clinic Platform.

This Week's Biggest Movers in AI

vs average of last 4 issues — click to explore

Entity	Now	Avg	Change
Agents	64	7.3	▲ +777%
OpenAI	63	7.3	▲ +763%
Funding	53	2.8	▲ +1793%
Chips	46	1	▲ +4500%
Regulation	45	2.8	▲ +1507%
AI Infrastructure	43	1.8	▲ +2289%
Anthropic	45	4.8	▲ +838%
Google	34	4.3	▲ +691%
Generative AI	27	1.5	▲ +1700%
NVIDIA	24	1	▲ +2300%
Open Source	18	1	▲ +1700%
Inference	17	0.3	▲ +5567%
Coding Tools	15	0.5	▲ +2900%
Cybersecurity	16	1.5	▲ +967%
Education	14	0.5	▲ +2700%