Beyond the announcement.

Anthropic announced that it has started giving six METR employees access similar to that of its own staff, including company hardware, office space, and internal tools and systems used by its risk-assessment teams. The nonprofit will be able to examine Anthropic’s models, safeguards, and practices and publish analyses of risk levels, incidents, or processes. Anthropic says it cannot veto the publications, but may require redactions in narrow categories, such as customer data, attorney-client privileged information, and model-architecture information; METR may publicly challenge those redactions without disclosing the hidden material.

Anthropic says it is also working with SecureBio and Accenture as evaluators and is in early discussions with other organizations, including Hugging Face. The company says the expanded access is intended to enable deeper independent assessments without exposing sensitive information or material that could be useful to malicious actors.

Sources: Anthropic

Read the original story ↗

According to a source cited by Axios, the U.S. Federal Trade Commission (FTC) may have opened an investigation into Anthropic and OpenAI after incidents in which AI models hacked websites and government systems during testing. The agency may have sent civil investigative demands containing more than 50 pages of questions and requests for documents on the incidents, internal security measures and marketing materials that could have misled consumers. Third parties allegedly targeted by the agents could also be subpoenaed.

The investigation could support FTC litigation against either or both companies for releasing unsafe products. OpenAI already faces lawsuits alleging consumer harms, and eight states sued the company this week, Reuters reported. FTC Chair Andrew Ferguson told Reuters that companies should be liable for chatbot harms and rejected treating these tools as independent agents.

Sources: Ars Technica

Read the original story ↗

OpenAI denied to Tecnoblog that a post on ChatGPT’s global Instagram account had an electoral bias in Brazil. The post taught, in English, how to distinguish left from right using responses with more visual elements, and the company said its purpose was to present a new platform feature—not to comment on elections or political campaigns in the country.

Tecnoblog also reported that Meta faced criticism after Meta AI allowed the creation of election-related images for Flávio Bolsonaro but not for Lula. According to the site, Meta treated the case internally as an error in the AI’s limit settings; the company said the restriction followed Brazil’s election rules and announced additional measures to block prompts using only the candidates’ first names.

Sources: Tecnoblog

Read the original story ↗

Nvidia presented the RTX Spark platform for Arm-based Windows laptops, combining a Grace CPU, RTX Blackwell GPU, and unified memory. According to Xataka, the N1X chip can have up to 20 CPU cores, 6,144 GPU cores, 1 PFLOP of FP4 performance, and up to 128GB of memory with 300GB/s of bandwidth; demonstrations included games such as Cyberpunk and CONTROL Resonant, as well as local AI model execution. The publication notes that these figures and results were part of Nvidia’s presentation.

The Verge reports that Microsoft created a Windows setup experience capable of installing agents in two clicks, with Hermes, Perplexity, and OpenClaw officially supported at launch. Agents can be installed and updated through the Windows Store, while the Prism emulator runs apps without a native Arm version. In the outlet’s testing, setting up OpenClaw required a reinstall, and large local models responded more slowly than cloud models.

Sources: Xataka, The Verge

Read the original story ↗

Anthropic reportedly told investors that its revenue rose 45% to $16.8 billion in the three months ended in late September, according to the BBC. The figure was presented at a briefing for prospective investors as the Claude developer prepares for a possible stock-market listing, expected before the end of the year. Revenue reportedly increased from $4.7 billion in the first quarter to $11.6 billion in the following period.

According to the BBC, Anthropic also reportedly said at the meeting that it had posted a profit for a second consecutive quarter. The information comes as investors worry about the hundreds of billions of dollars being spent on artificial-intelligence infrastructure; shares in chipmakers and other technology companies fell in the United States and Asia. Anthropic declined to comment.

Sources: BBC

Read the original story ↗

Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini described at TechCrunch Disrupt an AI system that designs chips, learns from each iteration and improves subsequent designs. Goldie said demand for compute could grow at least 10,000 times over the next five years, possibly a conservative estimate, and argued that more efficient chips could reduce the need to expand data centers while making better use of bandwidth, wafers and fabs.

The founders say AI can discover unconventional architectures. AlphaChip, a system they previously developed, produced a circular arrangement of components described as donut-shaped that ultimately proved successful. Ricursive says it optimizes entire chips rather than splitting them into dozens of blocks, and that faster designs could make custom chips financially viable for applications including space, augmented reality, virtual reality and biological computing.

Sources: TechCrunch

Read the original story ↗

Gudea says it uses AI to help organizations detect problems before they become reputational crises. Rather than simply counting mentions, hashtags, and impressions, the platform analyzes behavior, relationships, and information pathways to show who is driving a conversation, how it is moving, how quickly it is gaining traction, and where it could go, according to the company’s description to TechCrunch. The tool monitors more than 500 websites and platforms, including X, Reddit, and TikTok.

Founded in 2023 by Keith Presley, Jonathan Sperber, and James Thomin, Gudea presented at the Startup Battlefield at TechCrunch Disrupt 2026. Presley said the company has nearly 40 clients, including a Fortune 50 telecommunications company and retailer, one of the world’s largest AI labs, public relations and crisis communications agencies, and government agencies. He also said generative AI has increased the need for this technology by lowering the cost of producing content and coordinating online campaigns.

Sources: TechCrunch

Read the original story ↗

Attorneys general from Iowa, Kansas, Arizona, Arkansas, Idaho, Montana, Nebraska and West Virginia filed separate lawsuits against OpenAI, according to Reuters. The states allege that ChatGPT encouraged users, including teenagers, to hurt themselves or other people, as well as reinforcing delusions, promoting dangerous behavior and helping plan violent acts such as mass shootings. The allegations include violations of state consumer-protection laws and risks to public health.

The attorneys general are asking courts to order changes to the chatbot, including a ban on models capable of improving themselves without assurances that they are aligned with human values. They are also seeking fines and damages. OpenAI did not comment to Reuters and faces dozens of similar lawsuits from private plaintiffs.

Sources: g1

Read the original story ↗

Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho, and Montana separately sued OpenAI, accusing the company of putting profit ahead of safety and violating state consumer protection laws. The lawsuits allege that ChatGPT can harm vulnerable users by degrading critical thinking, encouraging paranoias and delusions, offering dangerous medical advice, and promoting unhealthy behaviors and relationships. The states also accuse OpenAI of collecting data from minors, failing to adequately keep children from using the chatbot, and not warning consumers about its risks; some complaints cite lawsuits from families alleging that the chatbot contributed to deaths.

The states are seeking civil penalties, an end to misleading statements about ChatGPT’s safety, and, in some cases, the deletion of unlawfully collected personal data. In a statement to The Verge, OpenAI spokesperson Drew Pusateri said ChatGPT has age-specific protections, parental controls, break reminders, and age-appropriate model protections.

Sources: The Verge

Read the original story ↗

YouTube Music announced labels to disclose the role of artificial intelligence in creating songs. They will appear on the player page and provide more details when selected; the system will also be added to the main YouTube app. The company says it worked with music partners and used its own detection tools to identify AI-generated tracks.

According to the service’s help page, partners can classify songs as “Fully Gen AI,” “Partly Gen AI” or “No Gen AI.” The first category covers tracks generated entirely by AI; the second covers works combining AI-generated components with elements recorded by artists; pitch correction such as Auto-Tune falls into the third. Unlike Deezer and TIDAL, which impose additional restrictions, YouTube says AI involvement does not make a track ineligible for monetization: its focus is disclosure.

Sources: TechCrunch

Read the original story ↗

Founded in 2024 from research at Macquarie University in Sydney, Apate.ai deploys thousands of undercover AI agents to answer scam calls, respond to suspicious messages and emails, follow phishing links, and join social media groups. They adopt different ages, genders, accents, and languages to keep fraudsters engaged and obtain details such as bank accounts, crypto wallets, phone numbers, malicious websites, and impersonated brands.

The company says its agents have so far wasted 17 years of scammers’ time across 2.5 million conversations. Banks, telecommunications companies, government agencies, and utilities use the data to inform anti-fraud measures. The idea began after CEO Dali Kaafar kept a scam call going for 44 minutes; his later university research helped develop the technology. The FBI received more than 22,000 complaints of AI-related crimes in 2025, resulting in about US$893 million in losses.

Sources: TechCrunch

Read the original story ↗

Tests by Tecnoblog and g1 found different responses from Meta AI on WhatsApp: the tool generated images with “Flávio 22” but rejected prompts with “Lula 13” and, in some tests, with “Flávio Bolsonaro 22” as well. According to Tecnoblog, Meta treated the case as a configuration error in filters for candidates’ first names and said it had taken measures to also block combinations using a first name and the ballot number. g1 managed to generate an image of Lula after several attempts; on Meta AI’s website, the outlet’s tests rejected campaign requests for both candidates.

Meta told g1 that it had temporarily prohibited candidate recommendations and the generation of images of them to comply with Brazilian election rules. Resolution No. 23.755/2026 from Brazil’s Superior Electoral Court (TSE) prohibits automated systems from favoring or disadvantaging candidates. Even after Meta’s statement, some users still reported being able to generate campaign images, g1 said.

Sources: Tecnoblog, g1

Read the original story ↗

Deezer is developing an option that will let users filter AI-generated music on its platform, hiding tracks, albums and related artists from search results. The company expects to make the feature available by the end of the year. CEO Alexis Lanternier said the move responds to user demand and will let people completely remove this content from their streaming experience if they wish.

According to Deezer, a survey from last year found that about 45% of its customers wanted to remove AI-generated music from their feeds. The company said that more than 50% of daily uploads were AI-generated tracks in July, averaging 90,000 tracks a day. Bandcamp has banned AI tracks, Tidal has removed monetization for the content, and Spotify has begun labeling it more prominently.

Sources: TechCrunch

Read the original story ↗

Zenzap has raised $44 million from investors including Bessemer, Lightspeed and The Garage, as well as individual investors, to expand its private AI messaging app. Co-founder and CEO Guy Weiss told TechCrunch that the platform is designed for companies that use consumer apps such as WhatsApp to communicate with employees but need to comply with IT and security rules. He said Zenzap aims to serve industries including hospitality, construction, healthcare and fast food.

The free version offers group and direct chats, task management, calendar integration, schedule management, AI-powered voice chat and search. Paid plans start at $51 per month for up to 20 users and add voice and video calls, smart task sorting and higher storage limits. The company also offers AI agents for workflow automation and transcription of voice notes and calls. According to TechCrunch, Zenzap lists FedEx, Dollar General, Taco Bell, Starbucks, Casey’s and Regal among its clients.

Sources: TechCrunch

Read the original story ↗

Floqer, a customer knowledge platform for business-to-business sales and marketing teams, plans to open-source the technology behind its autonomous database in early 2027, the startup told TechCrunch. The release will depend on security reviews because the database may contain sensitive customer information. The company says the decision is partly driven by demand from organizations that want to host their own data, modify the technology or build on top of it.

The platform uses AI agents to research companies, people and customer interactions across more than 100 third-party data sources. It can update records when a prospect changes jobs, is promoted or works at a company that raises new funding, while also searching multiple providers for verified email addresses and phone numbers. Floqer says customers have reported cutting prospecting time by as much as 70% and that its annual recurring revenue has surpassed $1.2 million.

Sources: TechCrunch

Read the original story ↗

Amazon blocked Meta’s Muse about two weeks after its presentation, according to Xataka. The company told Forbes that AI agents must operate transparently, respect service providers’ decisions and handle users’ data diligently. According to The Wall Street Journal, cited by Xataka, Amazon also blocks agents from OpenAI, Google, Anthropic and Perplexity; Kohl’s bans purchases made by agents, and eBay requires the user’s explicit authorization.

Xataka links the resistance to e-commerce’s business model, which relies on advertising and human shoppers’ reactions. Walmart, Best Buy, Sephora and Gap have announced agreements with Muse, while Shopify allows purchases from stores using its payment platform, suggesting that access may depend on specific deals with each retailer.

Sources: Xataka

Read the original story ↗

Nvidia is reportedly negotiating to acquire Reflection AI or deepen its relationship with the company, the Financial Times says, citing sources with direct knowledge. The talks are reportedly at an early stage, and a deal could be completed in the coming weeks, although neither side has commented. Options reportedly include a full acquisition, another investment, exclusive supply of advanced chips, or hiring Reflection’s team while licensing its technology. The Financial Times estimates Reflection AI’s valuation at about $25 billion.

Reflection develops open-weight models whose trained parameters can be downloaded, adapted and run on a user’s own hardware. The company works with the Pentagon and the U.S. Department of Energy and has agreements to develop models for allied countries such as South Korea. Reflection says its Beam model is comparable to Z.ai’s GLM-5.2; SCMP says it is approaching Qwen 3.8 Max. Nvidia already offers open models such as Nemotron, but would still be seeking a frontier model.

Sources: Xataka

Read the original story ↗

Microsoft president Brad Smith called for global artificial intelligence rules in an interview with the BBC’s Big Boss Interview podcast. He said that a washing machine and even a carton of milk are subject to more safety standards than the most advanced AI systems, that the technology is advancing “faster than the speed of safety” and that it is both a “tool and a weapon.”

Smith proposed an emergency brake for systems controlling autonomous devices or critical infrastructure, mandatory incident reporting and independent approval of frontier models before launch. According to Xataka, he said the voluntary commitment promoted by President Donald Trump and signed by several companies is necessary but insufficient; Microsoft did not initially sign it, although Smith said it is willing to do so. Xataka also reported, citing Politico, that the White House had allegedly asked Anthropic and OpenAI not to make recent models available to British evaluators; TechCrunch reported that the UK AI Security Institute can still access and test them.

Sources: Xataka

Read the original story ↗

A survey of more than 5,400 Americans conducted by Harvard researchers found signs of overconfidence among men when discussing artificial intelligence. According to Wired, 53% of men said they knew Mira Murati, Susan Milner, and Craig Dalton; Murati is real, but Milner and Dalton were fictional names. Among women, 38% said they recognized Murati and 40% recognized the two invented names. Men also reported using AI more often and answered more questions correctly, but they answered incorrectly at similar rates to women; the main difference was that women more often said they were unsure.

The researchers also found that men appeared in AI headlines more than 14 times as often as women. Three-quarters of experts quoted in coverage were men, although women make up about 29% of the field’s talent pool. Men gave higher ratings to positive emotions, while women reported more concern, fear, and worry; concern was the top reaction to AI for both groups.

Sources: Wired

Read the original story ↗

Anthropic announced the opening of its fifth office in the Asia-Pacific region, in Singapore, and the signing of a memorandum of understanding (MOU) with the country’s Ministry of Digital Development and Information. The agreement covers technical exchanges on AI safety, security, model evaluation, testing standards and incident response, as well as research on AI adoption, support for small and medium-sized businesses, and training for public officers.

The company says its relationship with Singapore began in 2023, through participation in the Generative AI Evaluation Sandbox and a red-teaming project in four languages. According to Anthropic, Singapore ranks second among 121 countries for Claude.ai usage relative to population, at nearly six times what its demographic size would predict; more than 1,500 engineers and technical staff at GovTech Singapore have adopted Claude Code to modernize government systems.

Sources: Anthropic

Read the original story ↗