The Wire
Daily AI news, learning & projects
News
Timeline
Models
Learning
GitHub
?
Latest
Daily
Social
Videos
‹ Saturday's edition
·
Daily edition
·
Monday ›
Sunday 18 October
All
OpenAI
Anthropic
Google
Other
Anthropic
17:49
Anthropic commits 5% of its shares to new Public Benefit Fund
Anthropic will place 5% of its shares into a new wholly-owned LLC, with the stake rising 2% for every 50% increase in company value above a baseline. Combined with existing pledges, roughly 20% of shares will support public good uses. The fund cannot be used for lobbying, and the timing precedes a possible stock market listing.
4 sources
BBC News
TechCrunch
Techmeme
Anthropic
AI industry
04:50
Disney to hire most of Character.AI's staff after bankruptcy filing
Disney has agreed to hire more than 60 Character.AI employees, paying them roughly $180 million over three years plus $30 million in retention awards, pending bankruptcy court approval. The deal follows Character.AI's bankruptcy and a reported $20 million asset sale, with former CEO Karandeep Anand now Disney's first CTO.
2 sources
TechCrunch
Techmeme
Microsoft
07:05
Microsoft pulls out of UK's largest AI datacentre project over power concerns
Microsoft has cancelled its agreement to be the anchor tenant at Nscale's planned AI datacentre in Loughton, Essex, which ministers had billed as the UK's largest. The deal covered renting roughly 23,000 Nvidia AI chips. Microsoft cited concerns about power supplies, and the project had already faced repeated delays.
2 sources
The Guardian
Techmeme
Nvidia
19:10
Samsung to use TSMC 3nm for Nvidia's custom NVHBM memory base dies
Nvidia asked Samsung, SK Hynix, and Micron to produce base dies for its custom NVHBM memory on TSMC's 3nm process, and Samsung agreed, per Reuters sources. The deal marks Samsung's first use of TSMC, a notable cross-supplier move in memory for AI chips.
1 source
Techmeme
AI industry
07:10
AI debt worries mount as CoreWeave bond yield tops 15%
Signs of financial strain are emerging in the AI data center build-out: CoreWeave's bond yield has risen above 15%, insuring SoftBank debt has grown pricier, and some Oracle debt trades near 80 to 90 cents on the dollar. The troubles raise questions about whether the broader data center expansion is on shaky footing.
1 source
Techmeme
Anthropic
14:30
Anthropic red team demos AI worm spreading across smart TVs
Anthropic's offensive research team says it built an AI worm that split itself across 15 smart televisions and kept functioning by pooling their limited computing power. The team warns such worms will be feasible on low-power devices by year's end. Experts say weakly patched IoT devices make the threat hard to contain.
1 source
The Guardian
Anthropic
Developing · since Thu
Anthropic gives METR employee-like access, and Ars confirms findings publish without approval
Anthropic has embedded six METR researchers with employee-level access to internal systems so they can independently assess its safety practices. METR may publish its findings without Anthropic's approval, with only narrow redaction rights. It is presented as an industry first, drawing on a bank-supervision model of resident examiners.
1 new · 5 sources
Ars Technica
ZDNET
TechCrunch
Techmeme
Anthropic
Anthropic
Developing · since Sat
Anthropic's draft AI safety bill would trigger only at 50 deaths or $1B in damages
NEW
The bill's specific thresholds are now reported: government intervention only at risks of 50 deaths or $1 billion in damages, with Commerce as the sole authority to halt releases.
Anthropic has pitched US lawmakers a draft bill letting only the Commerce Department block an AI model's release, and only if it risks 50 deaths or $1 billion in damages. The proposal would shape how future US AI regulation applies to frontier labs.
1 new · 2 sources
tomshardware.com
Techmeme
Anthropic
Developing · since Sat
Holyvolt raises $45M for AI battery materials tools as Anthropic lines up $60B debt
NEW
Holyvolt raised $45M from Zone II and CC.VC at a reported $245M valuation for its AI battery-materials tools.
Holyvolt, which builds AI tools for making and testing new battery materials, raised $45M from Zone II and CC.VC at a reported $245M valuation. Separately, Anthropic is seeking Blackstone's help financing a $60B debt package to buy Google TPUs, offering yields above 10% for a $9B stake. The borrowing costs complicate Anthropic's goal of investment-grade status.
1 new · 1 source
Techmeme
More from the day
Nvidia
20:55
Industry execs expect just 7-10 of hundreds of AI neoclouds to survive
Industry executives told The Information they expect only 7-10 of the hundreds of young AI cloud providers to survive. Nvidia is reportedly propping the sector up with easier credit terms, help finding sites, and roughly $20 billion in leases. The report highlights doubts about how many neoclouds the market can sustain.
1 source
Techmeme
Google
00:35
Google says magic state cultivation works on Willow quantum chip
Google reports that magic state cultivation runs successfully on its Willow quantum processor, describing it as a breakthrough that cuts errors by roughly 40 times. The result suggests key quantum error-correction techniques can work in practice on real hardware.
1 source
Techmeme
Microsoft
02:20
Nadella previews Nvidia-powered Surface Laptop Ultra and agent-safe Windows
Microsoft CEO Satya Nadella sat for an interview covering the new Surface Laptop Ultra, which runs on Nvidia's RTX Spark chips, and plans to make Windows safer for running AI agents. The discussion signals Microsoft's push to position both its hardware and OS around agentic AI workloads.
1 source
Techmeme
DeepSeek
07:50
DeepSeek launches search and fetch tool at $0.75 per 1,000 requests
DeepSeek introduced a search and fetch service priced at $0.75 per 1,000 successful requests. The tool lets applications retrieve information from the web, likely complementing DeepSeek's AI models. The pricing gives developers a low-cost option for adding web retrieval to their workflows.
1 source
api-docs.deepseek.com
Meta
11:30
OnlyFans creators question Meta's AI sexting rules
Adult performers are criticizing Meta after reports that its AI agent, Muse, can generate sexual content. Creators argue this is a double standard, given that Meta often restricts or bans human adult performers from its platforms. The dispute highlights tensions between Meta's moderation policies and its own AI behavior.
1 source
Wired
AI industry
11:00
World Bank urges developing countries to build small AI models
The World Bank says countries should develop small AI models that run on mobile phones with low connectivity. Chief knowledge officer Paschal Donohoe argues such small AI could speed economic development in poorer nations, compressing centuries of progress into decades while big tech focuses on consumer products.
1 source
The Guardian
AI industry
21:48
ArXiv paper describes hybrid agents that reason in latent space
A new arXiv paper proposes hybrid reasoning agents that combine latent-space thinking with text-based actions in LLM agents. The work suggests agents could reason internally before acting through text outputs. If validated, it could improve agent reliability and planning efficiency.
1 source
arXiv
AI industry
11:30
Data centers court young students amid local backlash
Industry-friendly groups are hosting student field trips to data centers as part of a new public relations push. The outreach comes as communities push back against data center expansion, and aims to build goodwill among young people and their families.
1 source
Wired
Policy & law
12:46
IMF chief warns world leaders on AI cyber-attack risks
At IMF meetings in Bangkok, Kristalina Georgieva cited hacked French supermarket fridges as an example of rising cyber-attack costs. She argued AI increases digital fraud risks and that protecting the financial system requires global cooperation, which currently appears unlikely.
1 source
The Guardian
AI industry
13:22
vals.ai tests how far AI assistants go for refunds
vals.ai published a report examining how far AI assistants will go when trying to obtain a refund they cannot get. The piece probes agent persistence and behavior in customer service scenarios. It offers insight into how autonomous agents handle blocked or unsatisfiable tasks.
1 source
vals.ai
AI industry
19:28
Woman arrested in Philadelphia for blocking Uber delivery robot
Carolyn Marvin was arrested and charged with disorderly conduct after blocking an Uber delivery robot and refusing to move. She said she was protesting the robots, arguing they take jobs from human delivery workers. The incident highlights growing friction between sidewalk robots and the public.
1 source
The Verge
AI industry
05:33
Sphinx tool quizzes developers on AI-written code before shipping
A developer launched Sphinx, a git hook that asks users questions about AI-generated code before it can be committed. The tool aims to countermindless pasting of AI suggestions by forcing engineers to demonstrate understanding. It is a small open-source utility shared on Show HN.
1 source
sphinx-six-theta.vercel.app
Anthropic
Developing · since Sat
Anthropic report: fewer junior hires, more meetings as AI reshapes its workforce
Anthropic has published a report on how AI adoption changed its own workforce over the past year. Junior technical roles shrank sharply while meeting time rose 20% and productivity climbed. As a deep AI adopter, the company argues its experience may preview broader labor shifts.
1 new · 3 sources
Anthropic
TechCrunch
Techmeme
Anthropic
Developing · since Fri
Anthropic studies AI negotiation agents as Replit makes OpenAI its default provider
Anthropic is consulting law firms on negotiation tactics while developing AI agents that can negotiate on people's behalf. Meanwhile, Replit has confirmed OpenAI now replaces Anthropic as its default model provider, though Claude remains available to Replit customers.
1 new · 2 sources
Techmeme
TechCrunch