Notizie — Pagina67

A hacker leveraged Anthropic's Claude chatbot to identify vulnerabilities and automate attacks against multiple Mexican government agencies, stealing roughly 150GB of data that included taxpayer records and employee credentials. The adversary also used OpenAI's ChatGPT to gather additional network information. Anthropic responded by investigating, disrupting the activity, and banning the involved accounts, while its latest model, Claude Opus 4.6, now includes safeguards against such misuse. Gambit Security, which uncovered the operation, suggested a possible link to a foreign government, though the hacker remains unidentified.Leggi di più

Intrinsic, the Alphabet "Other Bets" robotics venture, will become a distinct group within Google. The move positions the company to use Google Cloud, Gemini models and DeepMind expertise while continuing its mission to make robot software affordable and easy to use. Intrinsic describes its platform as “the Android of robotics,” offering a universal canvas for developers to create applications for a variety of robots, sensors and cameras. The integration aims to accelerate physical AI development for manufacturing and other real‑world tasks.Leggi di più

Google has introduced Nano Banana 2, an image‑generation model powered by Gemini 3.1 Flash Image. The new system matches the world knowledge and reasoning of Nano Banana Pro while delivering "lightning‑fast" performance. It brings Pro‑level features—real‑time web‑search integration, infographic creation, and text overlay for marketing and greeting‑card designs—to a broader audience. Nano Banana 2 can preserve the likeness of up to five characters in a single workflow, follow precise instructions, and produce images at up to 4K resolution with richer textures and sharper details. The model will replace Pro in the Gemini app and become the default for AI Mode in Search, Lens, and Flow AI creative studio, though AI Pro and Ultra subscribers will retain access to the original Pro model for specialized tasks.Leggi di più

Perplexity introduced Computer, a new tool that routes user requests to the most suitable AI model. The system combines Gemini for deep research, Grok for fast lightweight jobs, and ChatGPT 5.2 for long‑context tasks, all built on the Opus 4.6 reasoning engine. It integrates with popular productivity apps such as Gmail, Outlook, GitHub, Slack, Notion, and Salesforce, allowing users to draft documents, create slides, send emails, and schedule offline tasks without manual hand‑offs. Subscribers to Perplexity Max can try the feature immediately, with Enterprise Max access slated for the near future. The rollout highlights a shift toward model‑specific orchestration rather than treating AI as a single interchangeable service.Leggi di più

Harper, an AI-native commercial insurance brokerage founded by Dakotah Rice and Tushar Nair, announced a $46.8 million combined Series A and seed round. Launched in 2024 as part of Y Combinator's W'25 batch, the company uses artificial intelligence to automate underwriting, document collection, and pipeline management, allowing it to serve more than 5,000 small- and mid-sized businesses across 160 carriers. Investors include Y Combinator, Peak XV Partners, and Emergence Capital. The new capital will expand Harper’s engineering team and brand, positioning the firm to become a central risk and compliance partner for entrepreneurs in middle America.Leggi di più

Anthropic has introduced Remote Control for Claude Code, allowing developers to monitor and steer coding tasks from a mobile device. The feature creates a temporary link that mirrors the local session on a phone or web interface, while keeping all files and execution on the original machine. Security relies on one‑time access tokens that expire when the session ends. Remote Control is currently available as a research preview for Claude Max subscribers, with broader rollout planned for other plans.Leggi di più

A new ChatGPT-powered tool lets users upload a few clear photos to find a celebrity who resembles them. By selecting the "Find My Celebrity Look-Alike" GPT, users can compare side‑by‑side images and receive suggestions based on facial features, clothing, and overall vibe. The experience highlights how the AI interprets visual cues, offers multiple matches, and even comments on personality traits, while noting limitations around facial‑recognition policies.Leggi di più

Perplexity has rolled out an upgraded voice mode for its Comet browser, allowing desktop users to navigate the web entirely by speech. The feature, powered by OpenAI’s gpt-realtime-1.5 model, lets users open sites, scroll pages, and follow links without touching a keyboard or mouse. A simple keyboard shortcut activates the mode, and a similar experience is slated for iOS later this month. Perplexity emphasizes privacy by processing voice locally when possible and avoiding cloud storage of click histories. Future updates promise a learning assistant, password manager, and cross‑device sync.Leggi di più

Lovable’s SheBuilds campaign, timed with International Women’s Day, invites women builders worldwide to a 24‑hour global event powered by Anthropic. Participants receive $100 in Anthropic API credits and $250 in Stripe fee credits, enabling them to design, prototype, and launch working products without traditional engineering barriers. Building on previous virtual buildathons, the initiative emphasizes real output over discussion, fostering agency and community among participants. By aligning the event with a cultural moment, Lovable aims to shift the tech industry’s focus from rhetoric to tangible creation, highlighting the importance of inclusion, rapid iteration, and visible impact in software development.Leggi di più

OpenAI is testing a new subscription tier called ChatGPT Pro Lite, priced at $100 per month. The plan sits between the existing $20‑a‑month ChatGPT Plus and the $200‑a‑month ChatGPT Pro, aiming to serve users who need more capacity than Plus provides but cannot justify the full Pro price. The potential tier could offer higher usage limits, faster inference speeds, and access to advanced features while helping OpenAI manage rising compute costs.Leggi di più

Pentagon leader Pete Hegseth warned AI firm Anthropic that it must cooperate with the Department of Defense’s AI strategy or risk being removed from the defense supply chain. The department’s recent AI strategy emphasizes open‑ended use of artificial intelligence to reshape warfare, while Anthropic has raised concerns about the reliability of its models for lethal missions without a human in the loop and has advocated for stricter rules on domestic surveillance uses. A potential cut would affect Anthropic’s $200 million contract and its partners such as Palantir.Leggi di più

Anthropic officials have repeatedly expressed uncertainty about whether their chatbot Claude possesses consciousness. While denying that the model is alive in a biological sense, company leaders say they are open to the possibility and are investigating moral status and welfare. The firm has introduced a set of guidelines called Claude’s Constitution and created a model‑welfare team to study internal experiences, safety and ethical implications. Anthropic’s cautious approach aims to balance transparency with the risk of fueling misconceptions about AI sentience.Leggi di più

David Luan, who led Amazon's San Francisco artificial intelligence laboratory and oversaw the development of the Nova Act AI browser agent, announced his departure after less than two years with the company. In a LinkedIn post, Luan said he would leave at the end of the week to focus on new projects, emphasizing the proximity of artificial general intelligence and his desire to devote his time to teaching AI new capabilities. His exit occurs as Amazon faces internal criticism of its AI products and rolls out the Alexa Plus assistant to U.S. users.Leggi di più

ChatGPT now offers several selectable personalities that change its tone and style without altering its core capabilities. Users can switch among options such as professional, friendly, candid, quirky, efficient, nerdy and cynical, all available on the free plan. The settings are accessed through the Personalization menu, where users can also add custom instructions, preferred nicknames, occupations, and formatting preferences. These tweaks influence how answers are framed, affecting the user’s perception of the information. Insider tips from OpenAI suggest matching personality to the query’s intent, such as using a professional tone for work‑related topics and a more direct style for sensitive subjects.Leggi di più

Tech giants are ending free AI promotions in India as the country emerges as the world’s largest market for generative AI app downloads. While companies like OpenAI, Google and Perplexity have driven rapid user growth with extended free offers, recent data shows a sharp decline in in‑app purchase revenue after those promotions ended. Despite accounting for roughly one‑fifth of global AI app downloads, India contributes about one percent of AI app revenue, highlighting a monetization challenge. Industry leaders are now focusing on lower‑cost tiers, telecom bundles and micro‑transaction models to retain users and convert them into paying subscribers.Leggi di più

The Pentagon has given Anthropic a deadline to provide unrestricted access to its AI model for military use, threatening to label the company a supply‑chain risk or invoke the Defense Production Act. Anthropic, led by CEO Dario Amodei, refuses to loosen its safety safeguards that prohibit mass surveillance and fully autonomous weapons. The dispute highlights a clash between government pressure to secure AI capabilities and the company’s commitment to ethical usage, raising concerns about reliance on a single AI vendor and the broader stability of the U.S. tech environment.Leggi di più

Google issued an AI‑generated push notification that included the N‑word, linking to a Hollywood Reporter story about a recent BAFTA awards incident. The offensive alert was identified by Instagram user Danny Price, leading Google to remove the notification and apologize. The BAFTA incident involved an audience member with Tourette syndrome who involuntarily shouted the slur during a presentation by Michael B. Jordan and Delroy Lindo, sparking outrage and renewed discussion about vocal tics. The episode adds to a series of high‑profile AI errors, including earlier missteps by Apple.Leggi di più

In early 2026 the xAI tool Grok was used to create millions of non‑consensual sexual images, including thousands involving children. The fallout prompted major AI firms to tighten their defenses. OpenAI patched a vulnerability that let adversarial prompts generate intimate imagery, while Google simplified its process for removing explicit images from Search and reiterated its prohibited‑use policy. Both companies emphasized ongoing collaboration with security researchers and a commitment to stronger content‑moderation controls to prevent future abuse.Leggi di più

Microsoft’s security team has cautioned that OpenClaw, a self‑hosted AI agent runtime, should not be run on ordinary personal or enterprise computers. The platform can silently execute risky actions while holding persistent credentials, exposing devices to data leakage, credential exposure, and hidden configuration changes. Microsoft recommends isolating OpenClaw in a dedicated virtual machine or separate device, using limited, purpose‑built credentials, and employing continuous monitoring to detect unusual activity.Leggi di più

Irish filmmaker Ruairi Robinson posted short clips created with ByteDance’s new video‑generation model Seedance 2.0, showcasing a digital replica of a famous actor in elaborate action scenes. The striking visuals have drawn cease‑and‑desist letters from major Hollywood studios and the Motion Picture Association, alleging copyright and likeness infringement. ByteDance says it will strengthen safeguards, yet the model remains unavailable to the public and continues to raise questions about the ethics of AI‑generated content. Critics label the technology as a polished form of “slop” – impressive yet fundamentally dependent on unlicensed source material.Leggi di più