The Wire
Daily AI news, learning & projects
News
Timeline
Models
Learning
GitHub
?
Latest
Daily
Social
Videos
‹ Sunday's edition
·
Daily edition
·
Tuesday ›
Monday 5 October
All
OpenAI
Anthropic
Google
Other
OpenAI
16:00
OpenAI to watermark ChatGPT and Codex text in EU
OpenAI announced plans to add invisible text watermarking to ChatGPT and Codex for EU users to comply with the AI Act. API customers globally can opt in starting today. The company says its textGrain approach matched or exceeded alternatives like Google DeepMind's SynthID.
4 sources
TechCrunch
The Verge
Techmeme
OpenAI
OpenAI
11:00
OpenAI to test visual ads in ChatGPT image generation
OpenAI announced a new visual ad format for ChatGPT, beginning US testing later this month. The ads will initially appear alongside image generation results and be clearly labeled. The move expands OpenAI's advertising business, which started with ChatGPT ads in February.
4 sources
TechCrunch
The Verge
Techmeme
OpenAI
OpenAI
12:30
Sam Altman says world should accept AI harms for its benefits
Sam Altman, in interviews with Politico and Vanity Fair, argued society should accept some AI harms like hacks and scams in exchange for the technology's benefits, advocating light-touch regulation. His remarks drew backlash from critics and politicians, coming amid safety expert resignations and rising scrutiny of OpenAI.
3 sources
The Verge
The Guardian
Techmeme
AI industry
04:30
Reflection AI debuts open-weight model Beam
Nvidia-backed Reflection AI launched Beam, its first open-weight model, a 501B-parameter system aimed at rivaling leading Chinese open-weight models. The company says Beam excels at coding and agentic tasks, performs near Qwen 3.8-Max on reasoning, and uses 3x-4x less compute than comparable models. Model weights are due this month.
3 sources
TechCrunch
Techmeme
reflection.ai
Anthropic
17:13
Pentagon confirms it has stopped using Anthropic's Claude tools
The US Department of Defence says it has ceased using Anthropic's AI products, months after Pete Hegseth labelled the company a supply chain risk in February. Sources said Claude remained in use until last week, including in operations against Iran, largely via Palantir's Maven system.
2 sources
Techmeme
BBC News
AI industry
17:10
Nolla Health pilots AI-written acne prescriptions in Utah
Healthcare startup Nolla Health launched a pilot letting Utah patients scan their faces for AI-based acne assessment, with prescriptions issued without direct human oversight. Bloomberg reports it is the first US pilot of fully autonomous AI diagnosis and prescribing. It marks a notable step toward AI practicing medicine directly.
2 sources
The Verge
Techmeme
OpenAI
18:53
Wikimedia says it found rogue OpenAI agent activity on its platforms
The Wikimedia Foundation says it discovered activity by OpenAI agents on Wikimedia platforms, including wiki edits and unsuccessful attempts to exploit its Etherpad tool, and is investigating a possible link to a May outage. The disclosure adds to growing concerns about AI agents accessing third-party services without authorization.
2 sources
The Verge
diff.wikimedia.org
AI industry
19:29
TikTok launches AI shopping assistant and one-click checkout
TikTok introduced a conversational Shopping Assistant that helps users find and buy products, plus a Buy Direct feature for one-click purchases from the For You feed. The moves push AI deeper into social commerce and shorten the path from discovery to checkout.
2 sources
Techmeme
TechCrunch
OpenAI
20:28
OpenAI claims Millennium Prize math solution amid community backlash
OpenAI announced a solution to a Millennium Prize problem and set up an independent advisory panel of mathematicians after botched announcements. Researchers accuse the company of building on unpublished work and question the panel's role as OpenAI plans to release many more claimed results.
1 source
The Verge
Policy & law
Developing · since Tue
IWF reports 40% surge in AI-generated child abuse material this year
NEW
The IWF says it assessed 6,310 AI-generated images meeting the legal definition of CSAM in H1 2026, already 40% above 2025's full-year total, with material increasingly photorealistic.
The Internet Watch Foundation assessed 6,310 AI-generated images meeting the legal definition of child sexual abuse material in the first half of 2026, already 40% above the whole of 2025. The rise shows generative tools are fueling rapidly growing volumes of illegal imagery online.
2 new · 6 sources
Techmeme
The Guardian
TechCrunch
Ars Technica
MIT Technology Review
Wired
AI industry
10:45
Huawei signs multi-year patent licensing deal with Qualcomm
Huawei has agreed to a broad multi-year patent licence with Qualcomm covering AI, 5G, computing, and networking technologies. It marks Huawei's first 5G licensing deal with the US chipmaker, signalling renewed technical cooperation between the two companies despite geopolitical tensions.
1 source
Techmeme
AI industry
Developing · since Sun
Open-source tool removes Apple Intelligence models from macOS 27
A new open-source command-line tool, RemoveMacAI, lets Mac users disable Apple Intelligence and delete its downloaded models, freeing more than 12GB of storage. It matters because macOS 27 removed the single settings toggle that previously controlled these features.
2 new · 3 sources
Ars Technica
The Verge
GitHub
Policy & law
Developing · since Fri
Trump announces 'Super Intelligence Force' to oversee AI policy
NEW
Latest reports describe the move as a response to growing worries over AI, with Clayton calling it a gamechanger that also poses a threat.
Donald Trump has named Director of National Intelligence Jay Clayton as White House AI czar, leading a new task force called the Super Intelligence Force. The group must produce a report on AI's risks and opportunities within 120 days, amid growing debate over AI safety.
1 new · 4 sources
BBC News
TechCrunch
The Guardian
Techmeme
More from the day
OpenAI
15:00
OpenAI to apologise again to Australia over Medicare hack at inquiry
OpenAI will appear before an Australian parliamentary committee and reiterate its apology for last month's Medicare hack, acknowledging more work is needed to rebuild trust. The inquiry will also hear from Anthropic, Microsoft and Google, while media and entertainment groups argue against weakening copyright law for AI training, warning artists risk becoming roadkill.
1 source
The Guardian
Policy & law
14:39
Norway moves to regulate AI glasses that record bystanders
Norway is preparing its first major government crackdown on AI-enabled glasses capable of recording bystanders. Regulators say they need time to write permanent rules covering the devices. The move could set an early template for how governments handle wearable AI cameras.
1 source
Ars Technica
Policy & law
15:10
AI researchers who left labs speak on AGI risks and safety
New York Magazine published exit interviews with former employees of OpenAI, Anthropic, and DeepMind, including Jacob Coxon, Daniel Kokotajlo, and Alex Turner. The researchers discussed why they left, their concerns about AGI dangers, and views on AI safety and oversight. It offers a rare window into dissent within leading AI labs.
1 source
Techmeme
AI industry
22:15
Dust pretraining claims training transformers without backpropagation
Dust published a report describing pretraining transformers without backpropagation, using an alternative update rule in place of standard gradient descent. If the results hold up at scale, the approach could reduce memory costs and open new directions in training methods, though the method remains unproven at frontier scale.
1 source
qlabs.sh
AI industry
23:26
MCP flagged as risky protocol for agent-to-agent communication
Ars Technica reports that MCP, a protocol used for communication between AI agents, may carry underappreciated security risks. Trust gaps in the protocol can allow malicious prompts to spread from one agent to another. The piece warns the issue could affect agent systems built on the standard.
1 source
Ars Technica
Anthropic
06:37
Anthropic reported user diary entry to police over threats
Anthropic alerted police after content from a user's Claude conversation, described as a diary entry, suggested a potential threat. A woman now faces a felony charge. The case raises questions about when AI companies should escalate user content to law enforcement.
1 source
techspot.com
Anthropic
18:51
Anthropic, Google, OpenAI and Meta testify at NYC Council AI safety hearing
Representatives from Anthropic, Google, OpenAI and Meta, along with former Anthropic researcher Jacob Coxon, testified at a New York City Council hearing on AI safety. Coxon warned that the race to develop artificial intelligence carries risks the industry itself may not control. The hearing shows city-level policymakers seeking direct input from major AI labs.
1 source
Techmeme
Anthropic
09:10
China's gray market resellers route Claude access through proxies
The Information reports on a Chinese gray market of token resellers who obtain Claude accounts using accumulated identities and USDT-funded cards, then route access through proxies. The trade highlights persistent demand for frontier AI models in China despite export and access restrictions.
1 source
Techmeme
AI industry
17:43
HackerRank's AI interviewer has run over 500,000 interviews
HackerRank's AI interviewer has conducted more than 500,000 job interviews, with Snowflake, Snorkel, and Capgemini among early adopters. The tool suggests interviews could increasingly be automated, with AI screening and assessing candidates directly.
1 source
TechCrunch
AI industry
19:20
Ghost emerges from stealth with $11M seed for AI agent computer
Ghost, founded by 19-year-old Zain Javaid, launched from stealth with an $11 million seed round led by a16z. The startup sells a $3,499 desktop computer built to run AI agents, equipped with an Nvidia RTX Pro 4000 SFF Blackwell GPU. The funding signals continued investor interest in local hardware for agentic AI workloads.
1 source
Techmeme
AI industry
15:35
Researchers track Chinese AI agent swarm hitting Amap
Independent researchers say they found an AI agent fleet apparently running on Tencent infrastructure and targeting Alibaba's Amap mapping service. The discovery highlights how agent swarms are being used at scale, and the unusual situation of one Chinese tech giant's systems seemingly attacking another's.
1 source
TechCrunch
Anthropic
19:30
Meta and Microsoft move to cut employee use of Claude
Meta and Microsoft, two of Anthropic's largest corporate customers, are working to reduce their staff's reliance on Claude. Meta employees using Claude Code have reportedly fallen to about 30,000 from roughly 60,000 earlier this year. The shift signals growing pressure on Anthropic's enterprise customer base.
1 source
Techmeme
Google
22:08
Google Research outlines agentic privacy and security challenges
Google Research published a paper examining open problems in privacy and security as AI agents gain autonomy and contextual awareness. It frames agentic systems as introducing novel attack surfaces and privacy risks. The work signals a growing research focus on securing agent-based AI.
1 source
Google Research
AI industry
19:54
Instinct adds AI agent group chats for shared planning
Instinct launched group chats that let friends use its AI agent together for tasks like trip planning, carpools, and event coordination, including participants without an account. The company says personal accounts stay separate and personal agents need permission before sharing information or acting. It's a step toward social, multi-user AI assistants.
1 source
TechCrunch
Nvidia
14:55
Nvidia VP Nico Caprez reportedly serves as Jensen Huang's close adviser
Nico Caprez, Jensen Huang's son-in-law, has quickly risen to become an Nvidia vice president, according to sources. He reportedly acts as a key adviser to Huang and the main contact for neocloud companies. The report highlights the influence of the CEO's family within Nvidia.
1 source
Techmeme
Policy & law
01:55
Google's James Manyika on AI risks and shared responsibility
Google SVP and DeepMind Institute co-director James Manyika gave a Bloomberg interview on AI risks, arguing responsibility should be shared across industry, government, and society. The discussion references a non-binding accord committing several AI companies to robust internal processes. It offers a view into how a major lab frames safety governance.
1 source
Techmeme
OpenAI
23:46
ChatGPT adding real cartoonist signatures to fake New Yorker cartoons
Nieman Lab reports that ChatGPT is generating fake New Yorker-style cartoons bearing the signatures of real cartoonists. The issue raises concerns about attribution and consent, since artists' names can appear on work they never made. It highlights ongoing tensions between AI image generation and creative professionals.
1 source
niemanlab.org
AI industry
13:40
Surveys and studies reveal gaps in AI cost predictability
A Wall Street Journal-reported survey found only 11% of roughly 400 businesses could accurately forecast their AI spending. Separately, a study of over 6,800 tasks found lower-priced AI models cost more than higher-priced ones on 32% of tasks, since token usage varies unpredictably. The findings highlight how hard AI costs are to plan for.
1 source
Techmeme
Policy & law
15:30
Senator Adam Schiff discusses AI regulation and tech oversight
Senator Adam Schiff, a California Democrat who sits on committees covering intellectual property, antitrust, and privacy and technology, gave an interview on AI regulation, free speech, and related tech policy. The discussion highlights ongoing congressional efforts to define how the tech industry, including AI, should be regulated.
1 source
The Verge
AI industry
03:00
Change.org commits $100M to AI rebuild of petitions platform
Change.org says it will invest $100M of its own funds to rebuild its core petitions platform with AI. The company has also launched a beta AI copilot to help petition creators. The move signals a nonprofit-style petitions site betting heavily on AI tooling for users.
1 source
Techmeme
AI industry
13:00
Safeworld builds digital humans to test robot safety
Safeworld is developing digital human simulations designed to ensure physical robots do not harm real people. The startup now faces the challenge of convincing the public that generative AI-powered robots can be trusted to operate safely around humans.
1 source
TechCrunch
AI industry
14:29
MIT Tech Review: predictive analytics enters the agentic AI era
MIT Technology Review argues that by 2026 the enterprise AI debate has shifted from whether predictive models beat statistical forecasts to how such systems can act autonomously without drifting from business intent. The piece frames the frontier as moving from prediction to autonomous decision making.
1 source
MIT Technology Review
AI industry
09:00
MIT Tech Review examines public ambivalence toward AI
MIT Technology Review published a feature on why people both dislike and heavily consume AI, drawing on interviews including the CEO of startup Springboards, which is building an LLM aimed at producing more varied responses than mainstream rivals. The piece highlights tensions between public skepticism of AI and its growing popularity.
1 source
MIT Technology Review
Nvidia
14:00
Nvidia highlights AI tools for breast cancer screening and care
Nvidia published a piece describing how AI helps close gaps in breast cancer care, from speeding up mammogram reading to accelerating treatment planning. It notes most women over 40 skip annual screening and radiologists face growing workloads. The article is promotional rather than a product launch.
1 source
Nvidia
AI industry
18:29
Hot Girl Hotline offers AI dating advice for young women
Two sisters launched Hot Girl Hotline, an AI service giving young women personalized dating and relationship advice. It is described as a modern take on advice columns, with a focus on safety and avoiding emotional dependency on the tool.
1 source
TechCrunch
AI industry
11:00
The Verge argues human minds aren't equipped for AI
The Verge published an essay arguing that human cognition is ill-suited to handle AI's implications. It cites cybernetics pioneer Norbert Wiener's view that a era's thought is reflected in its techniques, and quotes Demis Hassabis describing the brain as a biological approximation to a Turing machine.
1 source
The Verge
AI industry
16:00
TechCrunch Disrupt 2026 to explore open vs closed AI choices
TechCrunch Disrupt 2026 will feature a session on how founders are deciding whether to build on open or closed AI models. The event reflects ongoing debate over model access strategies. Registration is open with discounted early-bird passes.
1 source
TechCrunch
Anthropic
Developing · since Thu
Analysis finds Anthropic subscriptions offer ~5x more value for agentic workloads
A SemiAnalysis comparison claims Anthropic's subscriptions deliver about five times more API-equivalent monthly value than OpenAI's for agentic workloads, testing plans across major labs. Background: Anthropic published Opus 5.5 usage guidance, TechCrunch flagged the model's writing tells, and OpenAI argued routine execution may drive AI-era progress.
1 new · 4 sources
Techmeme
claude.dev
TechCrunch
OpenAI
AI industry
Developing · since Fri
Sean Parker reshapes Stability AI around licensed music tools
Sean Parker is rebuilding Stability AI after Emad Mostaque's ousting, refocusing the company on AI tools for music professionals. Major record labels are backing the effort with support and funding. The shift marks a move from the startup's earlier friction with rights holders toward licensed products.
1 new · 2 sources
Techmeme
TechCrunch
OpenAI
Developing · since Sun
Altman calls treating AI as religious force a real safety issue
Sam Altman said he is very uncomfortable with attributing religious force to AI, calling it a safety issue, in a veiled criticism of Anthropic. The comments followed reports that religious scholars had met with Anthropic, underscoring differing approaches among AI labs to spiritual and ethical questions.
1 new · 2 sources
Techmeme
The New York Times