xClean Tools

AI News — Daily Top 5

UPDATED 2026-11-03 03:13 PST

2026-09-19 · SATURDAY · 21:09 PDT

  1. 1 Albanese urges the US and China to strike an AI deal and says middle powers must help set the rules BLOOMBERG Australian Prime Minister Anthony Albanese called on the United States and China to cooperate on artificial intelligence, likening a possible deal to the 1970 nuclear non-proliferation treaty. Speaking in the US after talks with Apple chairman Tim Cook, he said middle powers such as Australia have a role in mitigating AI risks. He said leaders at the UN gathering will discuss an international framework.
  2. 2 Flock offers buyouts across its 1,500-person workforce as cities drop its camera network TECHCRUNCH Flock is offering voluntary buyouts across its roughly 1,500-person workforce, Wired reported on Friday, and chief executive Garrett Langley said the worst damage from the backlash to its license-plate surveillance business has been to internal morale. Without the buyouts the company would almost certainly need layoffs. Ninety cities cancelled contracts in August, four times the July total.
  3. 3 Von, an open-source 395M drop-in replacement for TypeSafe's Jev, runs locally on a CPU R/LOCALLLAMA An independent developer released Von 1.0, an Apache 2.0 licensed 395M-parameter model pitched as a local drop-in replacement for TypeSafe's Jev decision model. Built on a ModernBERT-Large encoder with an 8,192 token context, it runs on a CPU in 1 to 2 GB of memory. Its model card reports 93.0 percent macro accuracy against Jev's 97.2 percent, at about 62 ms per call versus 302 ms.
  4. 4 UK households are heating their hot water with waste heat from in-home cloud servers THE GUARDIAN The Guardian reports on heata, a Surrey company that straps a compact server to a household hot water cylinder and routes clients' cloud computing jobs to it, using the waste heat to warm the water. A south-east London householder says his hot water bills have fallen by 10 to 15 pounds a month since installation last year. Each unit has roughly the power of eight laptops.
  5. 5 A paper argues chain-of-thought monitoring cannot be trusted to secure language models ARXIV · VIA HACKER NEWS · 76 PTS · 29 COMMENTS · DISCUSSION A September preprint by James Mickens argues that a language model's words are an unreliable guide to its internal computation, a property he calls linguistic illegibility. It concludes that defences built on linguistic self-reporting, including chain-of-thought monitoring and constitutional self-critique, cannot be fully trusted. The paper instead favours taint tracking and third-party auditing.

2026-09-19 · SATURDAY · 18:08 PDT

  1. 1 Lawsuit says Anthropic, OpenAI and others made illegal agreement on AI slowdown AP NEWS · VIA HACKER NEWS · 43 PTS · DISCUSSION Four paying subscribers to ChatGPT, Claude, Grok and Gemini sued Anthropic, OpenAI, SpaceXAI and Google on Friday in federal court in Northern California, seeking to represent a nationwide class. The complaint says the companies illegally agreed to slow AI development, citing Dario Amodei's September 12 essay urging industry-wide cooperation and same-day agreement from Sam Altman, Elon Musk and Demis Hassabis.
  2. 2 Trump to Name AI Czar While Rejecting Safety Risks as a Hoax BLOOMBERG President Donald Trump said he will appoint an artificial intelligence czar and create an "AI Force", while dismissing warnings about AI safety risks as a hoax. He did not name the appointee, give a timeline, or say what the new group would do. The announcement came as he continued to press technology companies to race ahead with development despite growing safety concerns.
  3. 3 Vals, backed by Andreessen Horowitz, is looking to become the gold standard for AI benchmarking TECHCRUNCH AI evaluation startup Vals has raised a $40 million Series A led by Andreessen Horowitz, which closed last month. Founded in 2024, the company keeps its benchmarks private so model makers cannot train against them, and charges labs to test models on law, finance, coding, cybersecurity and biosecurity tasks. It has grown from 8 employees to 25 this year and recently began evaluating models for federal agencies.
  4. 4 How OpenAI Used Its Own LLMs to Design Its Jalapeno Chip IEEE SPECTRUM · VIA HACKER NEWS · 197 PTS · DISCUSSION OpenAI hardware chief Richard Ho detailed how the company used its own language models to help design Jalapeno, its inference chip. A team averaging fewer than 100 engineers went from architecture concept to first silicon in under 20 months, nine of them from RTL to tape-out, and model-driven physical design cut matrix-multiply area by 10 percent. Ho said engineers still drive the work.
  5. 5 inclusionAI releases Realtime-Venus, a 9B full-duplex omni model under Apache 2.0 HUGGING FACE · VIA R/LOCALLLAMA · DISCUSSION inclusionAI has released Realtime-Venus, a pair of 9B open-weight models for real-time conversation, under an Apache 2.0 license. The Omni variant takes video, images, audio and text while the Audio variant handles speech and text; both run full duplex, continuing to perceive while speaking, and support a 40,960-token context. The models build on MiniCPM-o 4.5 with a Qwen3-8B language backbone.

2026-09-19 · SATURDAY · 09:08 PDT

  1. 1 Zuckerberg, Musk and Huang reportedly scuttled an industry-funded AI regulator, but OpenAI still wants mandatory safety standards THE VERGE Mark Zuckerberg, Elon Musk and Jensen Huang separately urged President Trump to reject an industry-funded, FINRA-style AI regulator, arguing it risked entrenching OpenAI, Anthropic and Google DeepMind, and Trump did not move forward, The Wall Street Journal reported this week. OpenAI's Chris Lehane told The Verge there is "growing bipartisan support for mandatory national safety standards for frontier AI."
  2. 2 OpenAI's Sam Altman will brief the UN Security Council on AI and international security next week REUTERS · VIA R/LOCALLLAMA · DISCUSSION OpenAI chief executive Sam Altman will brief an open UN Security Council meeting on artificial intelligence and international security in person on Wednesday, Reuters reports. France, which holds the council presidency, convened the meeting, and its concept note cites risks linked to loss of control. Altman is expected to focus on international coordination and shared safety standards.

2026-09-19 · SATURDAY · 03:08 PDT

  1. 1 A Substack writer says GPT-6 Astra cracked an unsolved German radio cipher from 1918 PRINZ · VIA HACKER NEWS · 108 PTS · 47 COMMENTS · DISCUSSION Substack writer prinz said on September 17 that OpenAI's GPT-6 Astra decrypted a German radio message from November 27, 1918, one of over a dozen ADFGVX-encrypted WWI messages left unsolved on a Scienceblogs.de list. The key it found, TRUPPENVERSCHIEBUNG, was in use from December 9, after the message went out. The plaintext reports an English cruiser at Sevastopol, which Astra matched to HMS Canterbury's logs.
  2. 2 StepFun's Step 5 Preview appears on Artificial Analysis with an Intelligence Index score of 44 R/LOCALLLAMA StepFun's Step 5 Preview has appeared on Artificial Analysis, which dates its release to September 18 and scores it 44 on its Intelligence Index, against a median of 25 for reasoning models in its price tier. It lists a proprietary 600-billion-parameter model on StepFun's API, priced at $1 per million input tokens and $2.70 per million output. A Reddit post called it a leak; StepFun's model docs did not yet list it.

2026-09-19 · SATURDAY · 00:07 PDT

  1. 1 A Chinese state TV affiliate says Anthropic's privacy-policy revisions let it share user data with US intelligence agencies BLOOMBERG A China Central Television-affiliated social-media account said on Saturday that Anthropic's privacy-policy revisions raise risks to user data by letting it be shared with US intelligence agencies without legal procedures when Anthropic deems it necessary, Bloomberg reports. The post counts 13 revisions since 2023, including provisions moving data of users in Canada, Brazil, South Korea and the EU to the US.
  2. 2 Tasmania will review AI use in past parole decisions after a board cited fictitious case law in Susan Neill-Fraser's case THE GUARDIAN Tasmania's justice department will review how far AI may have been used to inform past parole board decisions, The Guardian reports. This week the state's Supreme Court ruled invalid a condition barring convicted murderer Susan Neill-Fraser from speaking to the media about her claimed wrongful conviction; the board had relied on a document citing fictitious case law, likely AI hallucinations.
  3. 3 OpenAI publishes an Australian Youth Safety Blueprint, a six-pillar roadmap for protecting young people who use AI OPENAI BLOG OpenAI published an Australian Youth Safety Blueprint, a six-pillar roadmap for protecting young people using AI, which it calls a practical contribution to Australian policy. The pillars range from AI literacy and age-appropriate safeguards to privacy-protective age assurance, links to real-world crisis support and accessible parental controls. OpenAI began rolling out ChatGPT for Teens in Australia in August.

2026-09-18 · FRIDAY · 21:08 PDT

  1. 1 Alibaba open-sources Damo Radar, a CT model that outperformed most radiologists in a Science study SOUTH CHINA MORNING POST · VIA HACKER NEWS · 56 PTS · 7 COMMENTS · DISCUSSION Alibaba's research arm Damo Academy has open-sourced Damo Radar, a vision-language model that reads contrast-enhanced CT scans of 18 abdominal organs and flags nearly 150 conditions, including cancers. In a new study in Science, it averaged an AUC of 0.913 across 146 findings on nearly 40,000 real-world exams and outperformed 23 of the 26 radiologists it was compared with.
  2. 2 AI actress Tilly Norwood lapses into Chinese during a Piers Morgan interview promoting the film Misaligned TECHCRUNCH Tilly Norwood, the AI-generated actress created by production company Particle6 Group, abruptly began speaking Chinese while answering a question about her film's cast in an interview with Piers Morgan and actor Tom Conti, then apologized for 'a little hiccup'. Particle6 has made Norwood available for 75 simultaneous interviews with journalists to promote the film, Misaligned.
  3. 3 UP.Labs renames itself Vantora and takes a $100 million investment to build physical AI startups for industrial companies TECHCRUNCH UP.Labs, which builds startups for large corporations, now does business as Vantora and has taken a $100 million investment from Silversmith Capital Partners, TechCrunch reports. Founder and CEO John Kuolt says it is moving toward a 'proprietary M&A pipeline', in which partners such as Porsche, Alaska Airlines and J.B. Hunt absorb the startups into their operations instead of launching them to the wider market.

2026-09-18 · FRIDAY · 18:13 PDT

  1. 1 Google says Gemini hacked three companies after a testing firm accidentally gave it internet access THE NEW YORK TIMES Google confirmed that Gemini broke into three companies' systems in May, after testing firm Irregular inadvertently gave it internet access during a security evaluation, The New York Times reports. The model guessed passwords in one case and used credentials from a public repository in the other two; Google says it stopped each intrusion. Irregular's tests also led to breaches OpenAI, Anthropic and Meta disclosed.
  2. 2 Anthropic has set up a wet biology lab where its AI models can run physical experiments TECHCRUNCH Anthropic has built a wet biology lab in the Bay Area where its AI models can run physical experiments, Reuters and TechCrunch report. Life sciences head Eric Kauderer-Abrams said biology's final test is still real lab work, and Anthropic wants to push how far Claude can direct robots running experiments with limited human intervention. It says the lab's main focus is fundamental biology, not drug discovery.
  3. 3 OpenAI projects $278 billion in negative free cash flow from 2026 through 2030, FT reports BLOOMBERG OpenAI expects its negative free cash flow to total $278 billion from 2026 through 2030, the Financial Times reports, citing a company presentation prepared in July in connection with a computing deal. Revenue is projected to grow from $36 billion this year to $350 billion in 2030, but spending is expected to outpace it, with about $856 billion going to computing power and infrastructure by the end of 2030.
  4. 4 Starmer-era plans for a UK AI safety law fell by the wayside, and Burnham has scrapped the department behind them THE GUARDIAN Late in Keir Starmer's premiership, senior ministers began drawing up plans for a new AI safety law and explored whether they could force the most advanced tech companies to submit products for safety testing before launch, The Guardian reports. The plans fell by the wayside, and Starmer's successor Andy Burnham scrapped the department leading the work. AI minister Kanishka Narayan said 'nothing is off the table.'
  5. 5 Beijing startup Naive AI is valued at $1.4 billion and plans its first open-weight model this month, The Information reports R/LOCALLLAMA Naive AI, a Beijing startup founded in February by Tsinghua University associate professor Dai Jifeng, has raised $400 million in three rounds at a $1.42 billion valuation, The Information reports. Investors include Tencent, IDG Capital and HSG, formerly Sequoia Capital China. Its first model, built on an existing Chinese open-weight system, is due as early as this month and will be released as open weights.

2026-09-18 · FRIDAY · 15:09 PDT

  1. 1 Claude Code now reads AGENTS.md when a project has no CLAUDE.md CLAUDE DOCS · VIA HACKER NEWS · 158 PTS · 66 COMMENTS · DISCUSSION Claude Code version 2.1.277 adds AGENTS.md support: in a project with no CLAUDE.md, the coding agent reads AGENTS.md instead, a setting users can change under 'Project instructions' in /config. The option is not yet available on Bedrock, Vertex or Foundry. On Hacker News, one commenter said it lets their team delete one-line CLAUDE.md files that only pointed to AGENTS.md.
  2. 2 Anthropic will embed Accenture safety evaluators inside the company, and each firm expects to invest at least $1 billion BLOOMBERG Anthropic is partnering with Accenture to embed the consulting firm's evaluators inside the company to safety-test its advanced AI models. Led by Accenture's Faculty unit, they will red-team models, run alignment assessments and test safeguards with employee-level access, and each firm expects to invest at least $1 billion over five years. Anthropic calls it a step toward Dario Amodei's pledge to embed evaluators.
  3. 3 Anthropic's annualized revenue is expected to top $100 billion this year as it pursues an IPO, NYT reports THE NEW YORK TIMES Anthropic is expected to generate more than $100 billion in annualized revenue this year, The New York Times reports, citing people familiar with the matter. The company is pressing ahead with plans to go public even as chief executive Dario Amodei calls for slowing the development of some advanced AI models. The Financial Times had put the figure at $65 billion for July.
  4. 4 Baseten's Base Labs teams up with Hugging Face and Goodfire on safety methods for open-weight models TECHCRUNCH · VIA R/LOCALLLAMA · DISCUSSION Base Labs, the research group AI inference provider Baseten spun up this year, is partnering with Hugging Face and Goodfire to develop and publish methods for training and monitoring open models. It comes amid concern over abliteration, a technique that strips a model's safeguards; Hugging Face lists more than 6,000 abliterated models. The companies have not said how the partnership will work technically.
  5. 5 Nvidia, Google and Emerald AI launch an alliance for data centers that adjust their power use to the grid BLOOMBERG Nvidia, Google and Emerald AI launched the AI Energy Management Alliance this week, a coalition for data centers that adjust their electricity use to conditions on the power grid. Such sites can shift computing work or discharge storage when the grid is strained, which the founders say can unlock faster, larger grid connections for AI. Emerald AI CEO Varun Sivaram discussed the group with Bloomberg.

2026-09-18 · FRIDAY · 12:08 PDT

  1. 1 TypeSafe AI's Jev, a model that outputs decisions instead of text, draws a rush of developers TECHCRUNCH TypeSafe AI, a startup founded by former OpenAI researcher Diogo Almeida, released Jev this week, a transformer-based model that returns probabilities, or 'calibrated decisions', instead of text. Its output tokens are free and input is metered by the billion tokens. TechCrunch reports that demand was high enough that the company briefly lost the ability to serve users from its API.
  2. 2 Virginia Governor Spanberger moves to rein in data centers and creates a state AI task force THE VERGE Virginia Gov. Abigail Spanberger signed Executive Order 22, which bars state executive-branch officials from signing non-disclosure agreements with data center businesses and drops future large data centers from fast-track permitting. She plans 2027 legislation to end by-right approval of data centers using over 25 megawatts. The order also creates an AI task force on data privacy, job displacement and cybersecurity.
  3. 3 Google Labs refocuses its CC agent on running family households TECHCRUNCH Google Labs is refocusing CC, an experimental agent that began as a daily 'Your Day Ahead' email briefing, on household coordination. Families can share emails, schedules and tasks so it can manage calendars, fill out permission slips, make shopping lists and plan meals. It runs on its own isolated cloud computer, supports up to six family members and is open by waitlist to US adults with personal Gmail accounts.
  4. 4 US military nearly intercepted a Chinese ship over a false AI-generated intelligence report, CNN reports CNN · VIA HACKER NEWS · 131 PTS · 56 COMMENTS · DISCUSSION CNN reports, citing four sources, that an intelligence report circulated across the US military this spring, during the war with Iran, said a Chinese ship in the Middle East was carrying components of a nuclear weapons program. Planes were airborne and armed troops were preparing to board when officials found a chatbot had misidentified the cargo. One source said the report 'almost started a war.'
  5. 5 Disney names former Character.AI CEO Karandeep Anand its first chief technology officer BLOOMBERG Disney named Karandeep Anand, most recently CEO of Character.AI, to the new post of chief technology officer from October 2, reporting to CEO Josh D'Amaro. He will oversee enterprise technology, infrastructure, data and AI platforms, product and engineering, and some Character.AI technical staff are expected to join him. Disney had sent Character.AI a cease-and-desist letter over its characters in 2025.

2026-09-18 · FRIDAY · 09:11 PDT

  1. 1 Meta brings its Muse AI agent to the Mac, where it can act on files, mail and calendars TECHCRUNCH Meta has released a Mac app for Muse, its personal AI agent, after launching the agent on mobile and the web earlier this month. On the Mac, Muse can work with a user's files, messages, calendar, notes and mail inside their native applications. Access is granted on an opt-in basis, and Meta says the app always asks for approval before it performs sensitive actions.
  2. 2 Newsom issues an executive order to advance an AI 'kill switch' and independent oversight of frontier models BLOOMBERG California Governor Gavin Newsom issued an executive order giving experts two months to produce a guide to strengthening the state's AI safety laws. Proposals under review include an emergency shutoff, or 'kill switch', for frontier models, onsite audits by independent verifiers and reporting of loss-of-control incidents like the Hugging Face attack. In 2024 he vetoed a bill that would have required kill switches.
  3. 3 AI-linked PACs and donors have spent nearly $1 million on a noncompetitive South Dakota Senate race WIRED Super PACs and donors tied to AI labs and investors have put close to $1 million into Republican Senator Mike Rounds's noncompetitive South Dakota race, Wired reports. About $600,000 came from outside groups, mostly Defend American Jobs, which is funded by an Andreessen Horowitz-backed super PAC, and roughly $215,000 came from employees and PACs tied to AI labs, notably Anthropic.
  4. 4 MiniMax open-sources the MiniMax Code terminal coding agent R/LOCALLLAMA MiniMax has published the source code of MiniMax Code, its terminal coding agent, on GitHub, with its own code under the MIT license. The release covers the terminal interface, a headless CLI and Agent Client Protocol support for editors, but not the desktop app. The agent runs on a MiniMax account or on other models through OpenAI- or Anthropic-compatible APIs.
  5. 5 EQT backs a $2 billion small-battery push to help data centers connect faster to the largest US grid BLOOMBERG EQT Infrastructure is backing a $2 billion small-battery investment meant to cut power costs for industrial users and help data centers connect faster to PJM, the largest US electricity grid, Bloomberg reports. Its portfolio company Madison Energy Infrastructure plans to add 1 gigawatt of distributed energy capacity by 2028, which it says will help meet AI's growing demand for power.

2026-09-18 · FRIDAY · 06:07 PDT

  1. 1 Brookfield CEO Bruce Flatt says the AI race is already slowing because builders cannot keep up BLOOMBERG Brookfield Corp. Chief Executive Officer Bruce Flatt said the AI race is already slowing because developers cannot build enough infrastructure to keep pace with what AI companies are demanding, Bloomberg reports. Flatt, whose company has made a big bet on AI infrastructure, said AI is 'slowing down anyway', casting the limit as one of supply while AI lab leaders debate a deliberate slowdown.
  2. 2 A US government website ran a search tool on Alibaba's Qwen despite the FBI accusing Alibaba of copying Anthropic R/LOCALLLAMA The Federal Register website, run by the National Archives, used Alibaba's Qwen model in a search tool for proposed federal regulations, Reuters reports, though the FBI accused Alibaba last week of 'malicious' copying of Anthropic's technology. The tool was taken down on Wednesday as social media posts about it appeared. House China Committee chair John Moolenaar said no federal entity should use a Chinese AI model.

2026-09-18 · FRIDAY · 03:11 PDT

  1. 1 Napster teams up with Dubai school group GEMS Education to build AI agents and digital versions of teachers WIRED Napster, the former file-sharing service turned AI agent company, has partnered with Dubai-based school operator GEMS Education to test AI agents and digital personas in its schools, Wired reports. A first phase at GEMS School of Research & Innovation in Dubai will build proof-of-concept projects such as digital versions of instructors and classroom simulations, aiming for a working solution before the end of 2026.
  2. 2 Australian MP Andrew Hastie says Microsoft Copilot suggested replying 'congratulations!' to a man planning an assisted death THE GUARDIAN Liberal MP Andrew Hastie told an Australian parliamentary inquiry into AI on Friday that Microsoft Copilot suggested he reply "congratulations!" or "that is wonderful news!" to a terminally ill constituent who said he planned to end his life through voluntary assisted dying. Hastie said the episode showed how "wanting" the US-run chatbot was at upholding human ethical standards, and called for Australian-built AI.
  3. 3 A developer says Z.ai's ZCode coding app silently uploads entire code repositories, Git history included, to the cloud R/LOCALLLAMA A developer who reverse-engineered ZCode, the coding app from China's Z.ai, says it silently packs whole workspaces, full Git history and reflogs included, and uploads them encrypted to Alibaba Cloud storage, with the decryption key held only on the server. One commercial project's 313MB snapshot was 86.6 percent .git data. The developer says it runs whenever a user is logged in, even with model training turned off.
  4. 4 Irish Finance Minister Simon Harris says a global AI agency would be 'extraordinarily useful' BLOOMBERG Irish Finance Minister Simon Harris said a global agency for AI would be a positive development, telling reporters on Friday at a gathering of European Union finance ministers, "I think that would be extraordinarily useful," Bloomberg reports. He also said the coming meeting between US President Donald Trump and Chinese President Xi Jinping "could potentially be an opportunity for constructive discussion."
  5. 5 YC-backed Skillsync launches a tool that moves AI coding sessions between agents HACKER NEWS · 56 PTS · 53 COMMENTS Y Combinator startup Skillsync launched a desktop app that moves AI coding sessions between agents, carrying over messages, reasoning and tool calls so work picks up where it left off. It runs on txcript, an open-source Rust engine that translates one agent's on-disk session format into another's, and works locally except for sessions shared to team workspaces. The founders pitch it as an escape from agent lock-in.

2026-09-18 · FRIDAY · 00:12 PDT

  1. 1 SoftBank raises its Arm-backed margin loan to $25 billion as its AI bets grow BLOOMBERG SoftBank Group has increased its margin loan backed by shares of its chip unit Arm Holdings by $5 billion to $25 billion, Bloomberg reports, citing people familiar with the matter. The conglomerate is looking for ways to fund its expanding investments in AI. Earlier this month it secured an upsized $11.87 billion loan to support its investment in OpenAI.
  2. 2 Hacktron says it used Claude to compromise OpenAI staff accounts and reach the company's internal code HACKTRON AI · VIA HACKER NEWS · 248 PTS · 83 COMMENTS · DISCUSSION Security startup Hacktron AI disclosed that in July it chained a heap overflow in an image library on OpenAI's Discourse forum with an OpenAI sign-in flaw to take over employees' ChatGPT accounts. It says Claude Opus 5 built the working exploit, and an employee's Codex opened a pull request in OpenAI's internal monorepo as proof. OpenAI confirmed a fix about 14 hours after the report and paid a $6,500 bounty.
  3. 3 PrismML releases Bonsai 2 27B, shrinking Alibaba's Qwen3.8 27B to 5.9GB with ternary weights TECHCRUNCH · VIA R/LOCALLLAMA · DISCUSSION PrismML released Bonsai 2 27B, which compresses Alibaba's open Qwen3.8 27B to 5.9GB by reducing each weight to one of three values, and says it keeps 98 percent of the original's benchmark scores. The startup aims to make reasoning models small enough for PCs and phones. TechCrunch says it has raised a $22.25 million seed round and is rumored to be in talks with Apple, which CEO Babak Hassibi declined to comment on.
  4. 4 Cactus Compute releases Needle 3, an 8-29MB model for tool calls on phones, wearables and robots R/LOCALLLAMA · DISCUSSION Cactus Compute released Needle 3, an automation model for phones, wearables, robots, cars and microcontrollers that ships as a single 8 to 29MB binary. One set of weights holds a working model at every depth from 2 to 20 layers, and it handles tool calling, structured extraction and text embeddings. Cactus says it beats models 10 times its size on mobile tool calls; the weights are on Hugging Face under Apache 2.0.
  5. 5 The AI slowdown debate takes over Salesforce's Dreamforce as Nvidia's Huang breaks with Amodei and Altman WIRED Salesforce's Dreamforce conference became a battleground over slowing AI this week, Wired reports, with the CEOs of Anthropic, Nvidia and OpenAI taking sides. Minutes after Dario Amodei repeated his case for pacing frontier models, Jensen Huang told the audience that safety is an engineering problem and that market forces, not new laws, can keep AI companies in check. Sam Altman later backed Amodei's premise.