Nachrichten — Seite17

Google unveiled Nano Banana 2 Lite, an AI image generator that produces results in about four seconds per prompt. The speed lets users treat the tool like a rapid sketchpad, reducing the need for painstaking prompt refinement. While the Lite version trails the full‑size Nano Banana 2 in fine detail, reviewers found its output comparable for most creative tasks, making it ideal for brainstorming and high‑volume production.Weiterlesen

OpenAI CEO Sam Altman is in early talks with the U.S. government to give the federal government a five‑percent ownership stake in OpenAI and other leading artificial‑intelligence companies. The proposal, reported by the Financial Times, aims to smooth political hurdles and share the upside of the AI boom with the public. Altman’s plan would extend to rivals such as Google, Anthropic, Meta and xAI, and could involve sovereign wealth funds like Alaska’s Permanent Fund. Any agreement would still need congressional approval.Weiterlesen

OpenAI is in early talks to give the U.S. government a 5 percent ownership stake in the company, a move aimed at easing tensions with the Trump administration and addressing public concerns over artificial intelligence. The proposal, first pitched to former President Donald Trump last year, could be worth roughly $42.6 billion based on the firm’s latest $852 billion valuation. Officials say similar stakes might be offered to other U.S. AI firms, though details remain unsettled. The plan arrives amid heightened government scrutiny of AI models, export controls and broader debates about how to capture the sector’s wealth for the public.Weiterlesen

Anthropic announced that its Fable 5 model is back online after the U.S. Commerce Department lifted export‑control restrictions that forced a suspension in early June. The company said the shutdown stemmed from a “misunderstanding” over potential jailbreak risks and national‑security concerns. With new safeguards in place, Anthropic is re‑enabling the model for customers while warning that future frontier AI releases could face similar government scrutiny.Weiterlesen

Anthropic announced the launch of Claude Sonnet 5, branding it as the company’s most agentic model to date. The new system can devise plans, operate browsers and terminals, and execute tasks with minimal user guidance. Sonnet 5 outperforms its predecessor, scoring 80.5% on the Agentic Coding benchmark versus 67% for Sonnet 4.6. The model is now the default for all Anthropic users—including those on the free tier—and is available across the platform’s various plans and products. The release marks a shift in the AI race from conversational chatbots toward autonomous assistants.Weiterlesen

Anthropic announced Wednesday that access to its flagship AI models, Claude Mythos 5 and Claude Fable 5, would be reinstated worldwide after the U.S. Department of Commerce lifted export controls. The company said the models will roll back onto major cloud platforms throughout the day, with limited usage caps for Pro, Max, Team and enterprise customers. Anthropic also highlighted its Project Glasswing initiative, which proposes a shared standard for evaluating and mitigating AI jailbreak risks. The move follows a two‑week dialogue with regulators aimed at aligning safety safeguards with U.S. policy.Weiterlesen

A United Nations‑appointed Independent International Scientific Panel on Artificial Intelligence says the speed of AI advances is leaving existing regulatory frameworks behind. In a preliminary report released ahead of a global dialogue in Geneva, the panel highlights both the life‑saving potential of the technology and a surge in harms ranging from deepfake abuse to cyber‑crime, calling for stronger independent evaluation, international cooperation and common standards.Weiterlesen

Together AI, the cloud platform that rents Nvidia GPU clusters for AI workloads, announced an $800 million Series C financing that lifts its valuation to $8.3 billion. The round was led by Aramco Ventures and included a slate of tech‑focused investors such as Vista Equity Partners, General Catalyst, Emergence Capital, Nvidia and others. The funding comes as the company reports over $1.15 billion in annual bookings and a growing roster of enterprise customers.Weiterlesen

The U.S. Commerce Department told Anthropic on June 12 to stop offering its flagship models to users outside the United States, citing fears that the technology could be weaponized by foreign adversaries. Anthropic complied, pulling all access because it could not filter traffic by geography. The move centers on the company’s Mythos model, which officials deem capable of uncovering software vulnerabilities faster than top human experts, and on the more publicly available Fable 5, which now carries strengthened safeguards before it can be redeployed.Weiterlesen

Venice AI, a privacy‑focused artificial‑intelligence platform, announced a $65 million Series A round that values the company at $1 billion, officially making it a unicorn. The San Francisco‑based startup, led by crypto veteran Erik Voorhees, offers users access to over 200 AI models while keeping all input encrypted and never storing data on its own servers. With more than 850,000 unique visitors, 3 million active users and an average of 1.7 million daily API calls, the firm already generates an annualized run‑rate of over $70 million. The round was led by Dragonfly and included participation from Coinbase Ventures and North Island Ventures.Weiterlesen

The White House announced that export restrictions on Anthropic’s two most advanced AI systems, Mythos 5 and Fable 5, have been removed. Commerce Secretary Howard Lutnick confirmed in a letter to co‑founder Tom Brown that a license is no longer required for any export, re‑export, or in‑country transfer of the models. The decision follows a deal that obligates Anthropic to bolster safeguards against jailbreaks and other security risks, addressing concerns that the models could be used to bypass safety limits, especially in cybersecurity contexts.Weiterlesen

Anthropic unveiled Sonnet 5, a new version of its Claude AI that targets high‑volume, agentic workloads. The model delivers performance comparable to the company's Opus line while cutting token prices by up to 25 percent. Starting September 1, customers will pay $3 per million input tokens and $15 per million output tokens, with even lower rates available beforehand. Sonnet 5 will be the default on all Claude plans, including the free tier, and is positioned to curb runaway spending by enterprise users running automated AI systems.Weiterlesen

The U.S. Department of Commerce announced on June 30 that it is removing export restrictions on Anthropic’s Mythos 5 and Fable 5 artificial‑intelligence models. The move clears the way for the company to restore customer access on July 1 after weeks of a blanket suspension that barred foreign nationals—even those working in the United States—from using the tools. The lift follows Anthropic’s agreement to proactively detect and mitigate security risks, a demand that stemmed from industry warnings that the models could be vulnerable to jailbreaks and misuse.Weiterlesen

Google DeepMind unveiled Nano Banana 2 Lite, the latest addition to its Gemini 3.1 family, promising image generation that is both quick and affordable. The model can turn a text prompt into a picture in roughly four seconds, a fraction of the time required by earlier Google models, and costs about $0.034 per 1,000 images via the API. While it sacrifices some detail—particularly with small text and complex infographics—Google says the trade‑off is ideal for rapid prototyping and idea exploration. The new offering is now available across Google’s ecosystem.Weiterlesen

Google has added a new feature to its NotebookLM research tool that lets Ultra and Pro subscribers turn uploaded sources into 60‑second vertical videos. The AI‑generated clips combine narrated summaries with custom images, mimicking the format of TikTok. The rollout is currently limited to English and to paid accounts, with free‑user support promised soon.Weiterlesen

The U.S. Commerce Department announced Wednesday that Anthropic can resume offering its Mythos and Fable models abroad after a June‑12 export‑control rule forced the company to suspend public access. Secretary of Commerce Howard Lutnick said the firm agreed to monitor security risks, cooperate on protocols and report malicious use, paving the way for a phased reopening to vetted customers.Weiterlesen

Anthropic announced Wednesday that the Department of Commerce has removed export restrictions on its consumer‑facing AI model, Claude Fable 5, allowing the company to restore worldwide access. The move follows weeks of negotiations with the Trump administration and comes after the firm rolled out new safety classifiers to block previously identified jailbreak techniques. Anthropic said it will also re‑enable the model on major cloud platforms and will keep working with the government on broader access and security standards.Weiterlesen

Three former DeepMind scientists who built the first AI to beat professional poker players have launched EquiLibre Technologies, a Prague‑based lab that applies reinforcement‑learning models to stock markets. Backed by Creandum in a record Series A round that valued the company at $500 million, EquiLibre now trades billions of dollars daily alongside Tower Research Capital, boasting a flawless record of positive months since its debut. The founders say the venture is about advancing AI, not finance, and they aim to squeeze more performance out of fewer chips as competition heats up.Weiterlesen

Anthropic announced the launch of Claude Sonnet 5, a midsize foundation model that adds stronger agentic capabilities while cutting costs. The new version can plan, use tools like browsers and terminals, and run autonomous tasks that previously required larger, pricier models. Priced at $2 per million input tokens and $10 per million output tokens through Aug. 31, Sonnet 5 undercuts Anthropic’s own Opus 4.8 and rivals from OpenAI and Google. The company says the model improves on reasoning, coding and knowledge‑work benchmarks and offers safer behavior, positioning it as the default for both free and Pro plans.Weiterlesen

Anthropic announced Claude Science on Tuesday, a new AI workbench that lets researchers conduct computational studies without juggling multiple databases, pipelines and tools. The platform runs existing Claude models, including Claude Opus 4.8, and adds a project‑manager‑style assistant, prebuilt toolkits for genomics, protein structure and chemistry, and a built‑in fact‑checker. Early adopters at the Allen Institute and UCSF report faster analyses and reproducible results. Claude Science is now in beta for Pro, Max, Team and Enterprise customers, with a credit program to support postdoctoral and graduate projects through the end of 2026.Weiterlesen