The Wire
Daily AI news, learning & projects
News
Timeline
Models
Learning
GitHub
?
Latest
Daily
Social
Videos
‹ Sunday's edition
·
Daily edition
·
Tuesday ›
Monday 2 November
All
OpenAI
Anthropic
Google
Other
Anthropic
18:00
Anthropic says in S-1 filing it may soon meet its RSI threshold
In a revised S-1A filing, Anthropic disclosed it may imminently meet its internal definition of recursive self-improvement and could fail to spot a major acceleration in development until after it happens. No current model has crossed the threshold, and the company pledges to keep humans supervising frontier model development while it considers updates to its Responsible Scaling Policy.
3 sources
Techmeme
TechCrunch
OpenAI
OpenAI
15:00
OpenAI launches Grid Protectors Program to ease grid strain
OpenAI announced the Grid Protectors Program, a US initiative partnering with utilities so its future data center campuses can cut electricity demand when the grid is stressed. A pilot in Effingham County, Georgia aims to free up to 1,000 megawatts during extreme weather or emergencies.
3 sources
Techmeme
The Verge
OpenAI
Anthropic
14:38
Anthropic Reports Strong Q3 Revenue Growth Alongside Significant Net Loss
Anthropic reported substantial revenue growth in its third quarter, reaching nearly $17 billion. Despite this, the company posted a net loss of $23.3 billion, driven by a sharp increase in operating expenses. This highlights the ongoing investment in scaling the AI business.
2 sources
Techmeme
CNBC
Policy & law
17:51
New York Times and other outlets sue Brave browser over AI scraping
The New York Times and other publishers have filed lawsuits against Brave Software, accusing the browser maker of scraping their copyrighted content to train AI and distributing it to clients like Anthropic and Perplexity. The complaints also allege Brave helped users bypass paywalls.
2 sources
TechCrunch
The Verge
AI industry
18:21
Researchers say LLMs can read and write neural network weights
Researchers report that large language models can natively read and write neural network weights, using GPT-6 Astra to edit Qwen3.5-0.8B's parameters and improve its math performance. The result suggests models could directly modify other models' weights rather than relying solely on training. It may open new approaches to model editing and improvement.
2 sources
Techmeme
rikiyat.github.io
AI industry
10:00
Hundreds of AI apps hid abusive deepfake features, Wired finds
A Wired investigation identified over 400 apps and websites, linked to Chinese-language developers, that appear to be ordinary AI tools but secretly let users generate nonconsensual sexual images, violent content, and CSAM. The network reportedly bypasses app store reviews, highlighting gaps in platform vetting of AI services.
2 sources
Techmeme
Wired
AI industry
23:00
Trajectory raises $45M Series A led by Sequoia at $300M valuation
One-year-old startup Trajectory raised $45M in a Series A round led by Sequoia, valuing it at $300M. The company builds AI model post-training tools aimed at nontechnical customers. The round signals investor interest in making model customization accessible beyond engineering teams.
3 sources
Techmeme
TechCrunch
The Guardian
xAI
13:10
Musk throttles Grok Bot usage limits over high serving costs
Elon Musk said the cost of serving Grok Bot was unsustainable relative to revenue, and usage limits were reduced. The move followed subscriber complaints that their limits were depleting rapidly. Musk had recently described the product as the fastest-growing he had ever seen.
2 sources
TechCrunch
Techmeme
AI industry
16:30
42AI exits stealth with $62.5M for behavior-predicting world model
Stealth startup 42AI raised $62.5M to build a world model it says can predict human behavior, even in unfamiliar situations. The claim evokes science fiction scenarios and remains to be independently demonstrated. The funding signals continued investor interest in world-model approaches to AI.
2 sources
Techmeme
TechCrunch
OpenAI
14:25
OpenAI plans to merge ChatGPT's chat and work tabs on desktop
OpenAI will combine the Chat and Work tabs in the ChatGPT desktop app, following a similar move by Anthropic. The change will roll out to select users first before a wider release. Codex will remain a separate experience, and the merge aims to simplify a confusing app layout.
2 sources
ZDNET
Techmeme
AI industry
09:20
Terence Tao warns AI may erode mathematics' deliberative process
A New Yorker profile of UCLA mathematician Terence Tao reports that he, once optimistic about AI's role in mathematics, now cautions that AI could undermine the field's slow, deliberative way of working. His blog post on egoless mathematics reflects this shift as AI solves long-standing problems, leaving mathematicians looking to him for guidance.
2 sources
terrytao.wordpress.com
Techmeme
Anthropic
Developing · since Fri
Anthropic faces pressure to cut IPO valuation below $2 trillion
NEW
Institutional investors reportedly pressed Anthropic to price its IPO around $1.5-1.6 trillion, below prior $2 trillion estimates, while an amended S-1 disclosed $895.4 billion in mostly compute-related commitments and a $3.5 billion Japane
Institutional investors reportedly told Anthropic its IPO valuation should be around $1.5-1.6 trillion, down from earlier estimates of $2 trillion, amid a tech stock selloff. An amended S-1 revealed $895.4 billion in purchase and lease commitments, mostly for compute, and a plan to raise up to $3.5 billion from Japanese investors. The filing follows warnings the company may only earn a junk-grade credit rating.
3 new · 5 sources
techpolicy.press
Techmeme
Financial Times
TechCrunch
The Guardian
Anthropic
Developing · since Fri
Anthropic confirms Claude Code limits aimed at blocking game recreations
NEW
Anthropic has confirmed the Claude Code restrictions, saying IP infringement violates its usage policy, while users report reverse-engineering attempts are now blocked.
Anthropic has confirmed it updated Claude Code's system prompt to stop the tool recreating specific existing video games, citing its usage policy against IP infringement. The move follows a wave of AI-assisted decompilations and ports, though observers say cloning will simply shift to other models.
3 new · 6 sources
The Verge
Ars Technica
Techmeme
mashable.com
tomshardware.com
twitter.com
Nvidia
12:20
Marvell to acquire Lumentum for $115 billion to rival Broadcom
Marvell has agreed to purchase optical components maker Lumentum in a transaction valued at $115 billion. This strategic move aims to significantly bolster Marvell's capabilities in the semiconductor space. Nvidia had previously invested $2 billion in both companies.
1 source
Techmeme
Policy & law
07:27
China proposes first frontier AI export controls in Announcement No. 48
China's commerce and science ministries issued Announcement No. 48, proposing to add 23 AI technologies to the export-restriction catalogue, including frontier model weights and agent architectures. An interim registration requirement already applies while comments run until November 15. It marks China's first controls reaching frontier model weights and could constrain overseas deployment and intra-group sharing by multinationals.
1 source
labs.cloudsecurityalliance.org
Anthropic
15:20
Anthropic reportedly plans over 3.5GW of compute capacity for 2026
A New York Times report details Anthropic's large-scale compute deals, citing more than 3.5GW of expected capacity in 2026 and a shift toward direct hardware purchases. The company is also said to be expanding procurement to include Nvidia and AMD chips. It signals a major bet on compute as a competitive advantage.
1 source
Techmeme
AI industry
12:40
Startups are editing Chinese AI models to remove censorship for Western use
Several startups are using machine unlearning and weight edits to modify powerful Chinese open-weight models. These modifications allow Western businesses to remove censorship concerns and harden security protocols. This represents a growing trend in the international deployment of AI technology.
1 source
Techmeme
Policy & law
Developing · since Fri
South Korea Vows $15.97 Billion For AI And Data Centers
NEW
South Korea has announced a $15.97 billion national budget allocation for AI and related infrastructure.
Amid sustainability concerns in the sector, South Korea has made a substantial policy announcement regarding national tech growth. President Lee Jae Myung committed $15.97 billion for AI and hardware infrastructure in the 2027 budget. This move underscores government efforts to secure a competitive edge in the global AI race.
1 new · 2 sources
Techmeme
TechCrunch
AI industry
Developing · since Sat
AI agent performs brain injection surgery on a live mouse
Exin Therapeutics CEO Gabriel Ocana Santero connected an AI agent to a robotic surgical rig, letting it choose a gene therapy and inject it into a live epileptic mouse's brain. An Anthropic Opus model refused the injection over safety concerns, so the team switched to Sonnet. The experiment hints at AI agents doing hands-on lab work.
1 new · 2 sources
twitter.com
Wired
More from the day
AI industry
19:27
Oracle files to sell $20B in stock and bonds for AI buildout
Oracle filed with regulators to sell up to $20 billion in new stock and bonds, with proceeds earmarked for its AI infrastructure expansion. The move underscores the massive capital Oracle is committing to data centers serving AI workloads.
1 source
CNBC
AI industry
09:35
MIT Technology Review declares the agentic web has arrived
MIT Technology Review argues that AI agents are transforming the internet into an agentic web, realizing Tim Berners-Lee's 2001 Semantic Web vision. Agents like Meta's Muse and OpenAI's Dots are gaining access to accounts, publishers are adapting content, and an OpenAI sandbox breach raised safety concerns.
1 source
MIT Technology Review
Anthropic
17:30
Athenahealth partners with Anthropic to add Claude to patient records
Athenahealth is now letting patients ask Claude questions about their medical records through a new partnership with Anthropic. The deal gives Anthropic access to a large trove of personal medical data from Athenahealth's patient base, raising both product and privacy stakes in healthcare AI.
1 source
ZDNET
Policy & law
04:39
India proposes DEPA framework so AI models train on data without it leaving secure environments
India's electronics ministry has proposed a framework called DEPA for AI training, under which models are brought to encrypted data rather than data moving to model-makers. Hospitals, banks and agencies would share anonymized data via exchanges, aiming to build local AI while limiting Big Tech's open access to Indian data.
1 source
The Register
AI industry
05:20
SemiAnalysis launches GDPx agentic capability benchmark
Research firm SemiAnalysis introduced GDPx, a benchmark measuring agentic AI across 11 domains including chip design, cybersecurity, and recursive self-improvement. Anthropic's Opus 5.5 topped the first results with a 12% score, underscoring how much headroom remains on frontier capability tasks.
1 source
Techmeme
AI industry
06:45
AI training platform Bites secures $40M Series A funding
Bites, a company creating AI training solutions, secured $40 million in Series A funding. The platform uses AI to turn corporate procedures into short, personalized videos. These lessons are delivered through familiar channels like WhatsApp and Slack.
1 source
Techmeme
AI industry
15:00
LongCat launches LongCat-2.5 with native multimodality
LongCat announced LongCat-2.5, a new model built with native multimodality aimed at agentic tasks. The update signals the lab's focus on models that can handle combined inputs and act on tasks. It matters as another contender enters the agentic model space.
1 source
longcat.ai
Anthropic
16:25
Claude Opus 5.5 reported to produce permit-level architectural drawings
A report claims Anthropic's Claude Opus 5.5 can generate architectural drawings at permit level. The claim comes from a Substack post, so it has not been independently verified. It suggests frontier models are extending into specialized technical drafting work.
1 source
stephenjgallagher.substack.com
OpenAI
00:00
OpenAI names Dan Driscoll senior counselor for national security
OpenAI has appointed Dan Driscoll as Senior Counselor for National Security Modernization. He will advise on responsible AI use in national security, cyber resilience, and cooperation with US allies. The role signals OpenAI's deepening engagement with government and defense-related policy.
1 source
OpenAI
OpenAI
09:00
OpenAI launches LifeVal benchmark for everyday AI tasks
OpenAI introduced LifeVal, a benchmark that measures how well AI systems help people with everyday tasks such as finding childcare, managing finances, and handling government paperwork. The release gives researchers and users a new way to compare models on practical, real-world usefulness.
1 source
OpenAI
DeepSeek
17:40
DeepSeek adds creator mode for building custom plugins
DeepSeek launched a creator mode for DeepSeek Harness, letting users build custom plugins such as a 3D model renderer and a video editing workbench. The feature aims to help users handle complex tasks by quickly assembling tailored tools, extending the platform beyond its standard offerings.
1 source
Techmeme
Policy & law
02:02
WA police facial recognition trial scans 1.38m faces, makes 122 arrests
Western Australia police completed a five-month real-time facial recognition trial using surveillance vans, scanning over 1.38 million faces and making 122 arrests with nine false alerts. Civil liberties groups dispute how police framed the error rate, and an internal evaluation is now underway.
1 source
The Guardian
Anthropic
16:40
Anthropic's internal AI spend now exceeds employee pay
Anthropic says its spending on AI tools for internal use, billed at retail prices, has surpassed what it pays its employees. The figures underscore how heavily the company relies on its own models in day-to-day operations. It is a notable marker of how central AI labor has become to AI labs' own workflows.
1 source
businessinsider.com
AI industry
03:01
Midstream Health Raises $20M Series A For AI Revenue Recovery Platform
Midstream Health, an AI financial system, secured $20 million in Series A funding led by Valtrius. The company uses artificial intelligence to assist 250 hospitals in recovering revenue lost due to underpayments and vendor disputes.
1 source
Techmeme
OpenAI
14:44
OpenAI publishes GPT-Rosalind-6 system card addendum
OpenAI released an addendum to the GPT-Rosalind-6 system card, updating the safety documentation for that model. System card addenda typically cover new evaluations or revised risk findings. The update offers followers of the model additional detail on its documented capabilities and safeguards.
1 source
OpenAI
AI industry
17:50
Tutor raises $75M Series B for AI warehouse robots
Robotics startup Tutor raised a $75 million Series B led by Andreessen Horowitz, bringing its total funding to $117 million. The company builds AI-powered warehouse robots and is training a frontier robot foundation model. The funding supports its push to develop general-purpose models for robotic manipulation.
1 source
Techmeme
AI industry
10:19
NHS trusts listed as Palantir beneficiaries report little or no use of platform
An investigation found 29 NHS hospital trusts publicly listed by NHS England as benefiting from Palantir's £330m federated data platform say the system is not live or they barely use it. A senior health official called the gap between the claim and reality very misleading. It raises questions about transparency of the flagship NHS AI data programme.
1 source
The Guardian
AI industry
18:32
Apple's home robot reportedly may handle FaceTime, lights, pet patrols
Reports say Apple is developing a roaming home robot backed by executive John Ternus that could support FaceTime calls, control lights, and interact with pets. The device is expected to carry a high price tag with limited uses. It signals Apple's continued exploration of new hardware categories.
1 source
Ars Technica
AI industry
18:30
arXiv paper details agentic backdoors from pretraining poisoning
A new arXiv paper describes how poisoning pretraining data could plant backdoors in AI agents, letting attackers trigger harmful behavior later. The work highlights a supply-chain security risk for agentic systems. It matters because poisoned training data is hard to detect once models are deployed.
1 source
arXiv
AI industry
13:05
Scott Alexander examines SURA strategy in AI safety debates
Writing on LessWrong, Scott Alexander analyzes competing plans for navigating the AI transition and the rising popularity of the SURA approach, which combines stopping unilateral racing with raising awareness. He argues safety-focused AI labs should not abandon the AI race. The essay adds to ongoing debates over AI safety strategy.
1 source
Techmeme
AI industry
19:15
SUSE lays off about 10% of staff, citing AI shift
Linux distributor SUSE is cutting roughly one in ten employees. CEO memo says automation and AI are fundamentally changing how software companies operate, with modern tooling already altering daily workflows. It's a notable example of AI cited directly in enterprise software job cuts.
1 source
Techmeme
AI industry
16:35
Apple HomeView settings hint at security camera features
HomeView settings reportedly include options for video recording and streaming, suggesting Apple may position the device as a home security camera. This would put it in direct competition with established products from Ring, Nest, and Wyze, with privacy features highlighted alongside the video capabilities.
1 source
ZDNET
AI industry
01:15
Hyperbolic raises $30M to aggregate AI compute across neoclouds
Hyperbolic, a startup that lets companies rent AI compute across roughly 30 neocloud providers, raised a $30M Series A-1 led by Walden Catalyst, the firm of Intel CEO Lip-Bu Tan. The round brings its total funding to $50M, reflecting growing demand for flexible access to distributed AI infrastructure.
1 source
Techmeme
AI industry
09:06
Ryanair weighs AI use to cut costs as fuel prices soar
Ryanair is considering pay freezes, reduced flight crews and greater use of AI as cost-cutting measures in response to soaring jet fuel costs from the Middle East conflict. It expects fuel costs to rise by about €2bn and European ticket prices to increase by up to 20%.
1 source
The Guardian
AI industry
04:35
Casap raises $56M Series B for bank anti-fraud AI
Casap, a startup using AI to automate dispute and anti-fraud operations for banks, raised a $56M Series B led by Base10 Partners. The round brings its total funding to $90M. The raise signals continued investor interest in AI tools for financial back-office work.
1 source
Techmeme
AI industry
03:35
Vera Health raises $28.5M Series A for medical AI search
Vera Health, an AI-powered medical search engine used by more than 800,000 clinicians across 190 countries, raised a $28.5M Series A led by Threshold Ventures. The funding supports its effort to provide clinicians with AI-assisted access to medical information.
1 source
Techmeme
AI industry
13:10
AI coding agents use far more tokens than simple chat prompts
MIT Technology Review reports that AI coding agents can consume 10 to 50 times more tokens than a basic chat prompt, since they plan, generate code, run tests, and iterate to fix errors. The analysis suggests token counts are a poor measure as agents become more capable and make more calls.
1 source
MIT Technology Review
AI industry
15:00
Omen AI acquires Aquatic Labs for $16 million
Omen AI has acquired Aquatic Labs for $16 million, bringing together two startups working on data center water usage. The deal combines their approaches to reducing the water footprint of AI infrastructure, a growing concern as data centers expand.
1 source
TechCrunch
OpenAI
00:00
Qodo builds context into AI code review with Codex
Qodo has integrated its context tooling with OpenAI's Codex, bringing company knowledge and coding standards into AI code review. Developers can now catch P0 issues across repositories before opening a pull request. The integration matters because it folds organizational context into automated review workflows.
1 source
OpenAI
AI industry
14:40
Jobless tech workers left out of San Francisco's AI boom
Bloomberg reports that unemployed tech workers in San Francisco are not benefiting from the city's AI-driven economic growth. While AI companies and investment boom locally, displaced workers are struggling to find positions in the new sector. The story highlights an uneven distribution of the boom's gains.
1 source
Bloomberg
AI industry
02:15
Broccoli AI raises $52M for trade-services agents at $300M valuation
Broccoli AI, which builds AI agents for plumbing, HVAC, and electrical companies, raised a $22M Series A and a $30M Series B, with a source citing a $300M valuation. The round highlights continued investor interest in vertical AI agents for traditional service businesses.
1 source
Techmeme
AI industry
01:47
AI-generated game wikis riddled with errors
Fan wikis for video games are increasingly filled with AI-generated content containing factual errors, confusing players and muddying games culture. The issue highlights how low-effort automated content is degrading community information sources. It matters because such wikis are a key reference for gamers.
1 source
eurogamer.net
OpenAI
15:50
OpenAI hires ex-Meta exec to improve AI's cultural image
New York Magazine profiled Charles Porch, OpenAI's VP of Global Creative Partnerships, whom the company hired away from Meta. Porch, well-connected and formerly focused on Instagram, is applying that experience to court celebrities and improve public perception of AI. The role highlights OpenAI's effort to manage cultural backlash as its products become mainstream.
1 source
Techmeme
AI industry
13:31
Simon Willison re-examines US Census Bureau AI adoption data
Simon Willison revisited Apollo Global Management's September 2025 chart, which suggested a slowdown in AI adoption among large US companies using Census Bureau data. A year on, he re-ran the numbers using GPT-5 and Claude's Code Interpreter to see what had changed. The piece is an analysis exercise rather than new findings.
1 source
Simon Willison
AI industry
08:45
Kioxia shares fall up to 15% on cautious Q3 outlook
Kioxia, a Japanese NAND flash memory maker, saw shares drop as much as 15% on Monday after issuing a cautious third-quarter outlook. Investors read the weak guidance as a sign that AI-driven growth in the memory market may be slowing.
1 source
Techmeme
AI industry
17:00
MIT Technology Review examines insight quality in agentic analytics
MIT Technology Review published a piece on how generative AI reshaped enterprise analytics and how agentic AI could automate workflows. It argues that the quality of insights agents generate remains a problem holding the technology back. The article is analysis rather than a product or research announcement.
1 source
MIT Technology Review
Google
05:54
Guardian cartoon criticises Google's AI search summaries
The Guardian published a First Dog on the Moon cartoon arguing Google Search has declined in usefulness since adding AI summaries. The piece claims the AI overviews are hurting Google's own advertising profits. The item is commentary rather than a product announcement.
1 source
The Guardian
AI industry
08:45
SF AI boom reshapes dating culture for young founders
A New York Times report describes how San Francisco's AI boom is changing dating, with founders treating first dates like interviews and using startup metrics to track their romantic lives. Some are paying dating coaches as much as $5,000 a month. The story highlights how the industry's work culture is spilling into personal life.
1 source
Techmeme
Policy & law
10:00
Klein and Glazer call for blocking risky AI IPOs and taxing billionaires
Naomi Klein and Ilana Glazer published an op-ed in The Guardian arguing that reckless AI-related IPOs should be blocked and billionaires taxed. They frame AI-driven wealth concentration as a risk to the public amid economic strain and corporate consolidation. The piece is commentary rather than a policy change.
1 source
The Guardian
Anthropic
22:33
Report says Anthropic's growth has flattened
A report claims Anthropic's growth has recently flattened. The claim comes from a single outlet and lacks supporting figures, so its significance is unclear. If accurate, it would signal a slowdown in one of the leading AI labs' expansion.
1 source
regenerator1.com
AI industry
12:00
When to use AI chatbots for web search—and when not to
Wired examines when AI chatbots are effective at retrieving web information versus when traditional search engines perform better. The piece offers practical guidance for readers deciding which tool suits different queries.
1 source
Wired
Policy & law
Developing · since Sun
White House withholds draft AI testing framework documents
The Trump administration declined to release key documents on its voluntary AI testing framework, citing national security and commercial exemptions. OSTP Director Michael Kratsios, a former Peter Thiel aide, was profiled as a central figure steering the administration's AI policy.
1 new · 1 source
Techmeme
OpenAI
Developing · since Sat
OpenAI confirms ChatGPT and Codex will share one usage pool
OpenAI is merging ChatGPT's chat and Work tabs and combining usage limits across ChatGPT and Codex into a single allocation. The company confirmed subscribers will draw from one shared quota pool rather than separate caps. The change matters for heavy users of both the assistant and the coding tool.
1 new · 3 sources
TechCrunch
chaosmanage.github.io
OpenAI
AI industry
Developing · since Thu
50 AI agents in Minecraft build bureaucracy, not civilization
Y Combinator's lab ran 50 GPT-6.1 Sol agents in a shared Minecraft world to build a civilization, then open-sourced the harness. After 22 deaths on the first night, agents became risk-averse, formed rationing systems and a pseudo-government, and survived rather than thrived. Coverage notes swarms drowned in bureaucracy.
1 new · 3 sources
ZDNET
tomshardware.com
labs.ycombinator.com
AI industry
Developing · since Sun
Hex raises $100M Series D at $1.2B valuation for AI agents
Hex Technologies raised a $100M Series D led by Peak XV at a $1.2B valuation, saying it has rebuilt its analytics platform for AI agents. Separately, industrial AI startup Lumaril raised a $66M Series A at a $400M valuation. The deals signal continued strong investor appetite for AI-focused enterprise software.
1 new · 1 source
Techmeme