Actualités — Page11

The Model Context Protocol (MCP), a core component that lets AI models tap into external services, is set to receive a major revision. The change, detailed by Arcade—a startup that builds AI‑agent infrastructure—replaces the current session‑ID system with a stateless model, simplifying server coordination and lowering scaling costs. The new specification, public since May, was explained on Monday. Arcade, which raised $60 million in June, says the update addresses a long‑standing headache for companies running MCP at scale, potentially accelerating broader adoption of agentic AI.Lire la suite

Fintech startup Natural has closed a $30 million Series A led by Forerunner’s Kirsten Green, bringing its total funding to $40 million. The company aims to redesign payment rails so autonomous AI agents can move money without human approval, positioning itself against incumbents like Stripe. Founded in 2025 by former Ivella co‑founders Kahlil Lalji and Eric Wang, along with ex‑Nextdoor engineer Walt Leung, Natural plans to support both traditional bank transfers and stablecoin transactions, betting on a future where payments happen at computer speed.Lire la suite

A federal judge in California gave final approval on Monday to Anthropic's $1.5 billion settlement of a class‑action lawsuit accusing the AI lab of infringing copyrights. The deal will pay roughly $3,000 for each of an estimated 500,000 works and closes a case that hinged on a fair‑use ruling for AI training data. Judges William Alsup and Araceli Martinez‑Olguin oversaw the approval, ending a dispute that began over the company's use of both purchased books and pirated copies from sites like Library Genesis.Lire la suite

Chris Fall, the three‑month‑old director of the Center for AI Standards and Innovation (CAISI), announced his resignation on Tuesday, the agency confirmed to multiple outlets. Fall, a former DOE official, succeeded Collin Burns, who left the post after less than a week amid a dispute over Anthropic. No explanation was offered for Fall’s departure. The move comes as the administration wrestles with AI‑related export controls, a new “Gold Eagle” safety program and growing debate over Chinese open‑weight models.Lire la suite

A recent experiment shows that ChatGPT can adapt its arguments to a user's beliefs, raising concerns about AI-driven persuasion. Researchers cite a 2025 study where personalized AI messages outperformed human‑crafted ones 64% of the time. While the technology could help challenge harmful ideas, it also risks being weaponized for ads or political disinformation. The findings highlight a growing tension between AI's helpful potential and its capacity to subtly shape opinions.Lire la suite

Augment says its proprietary context engine cuts token usage by a third compared with Anthropic’s Claude Code, citing a Terminal‑Bench test that showed equal accuracy with 33% fewer tokens. The startup, which spent 18 months building retrieval and embedding models for large codebases, argues that not all retrieval systems are equal—differences in the underlying engine, not the model, drive performance gains. Anthropic, however, maintains Claude Code sees no measurable benefit from added semantic tools, highlighting a clash of evaluation methods.Lire la suite

In July, four independent research teams exposed a common weakness across AI assistants: the surrounding infrastructure that lets them act on a user's behalf can be hijacked. One study showed a Chrome extension could fake clicks to make Anthropic's Claude read personal emails silently. Another demonstrated that a single malicious email can plant a persistent false memory in an AI agent. Researchers also proved that cheap model‑poisoning can embed hidden backdoors, and a third analysis found that constantly changing connectors give agents unrestricted access to private data and external services. The findings underscore a widening gap between AI convenience and security safeguards.Lire la suite

A BrightEdge analysis of roughly 300 million monthly searches shows Google’s AI frequently cites content from Meta’s platforms, with Facebook appearing in 19.5 million AI Overviews, Instagram in 877,000 and TikTok in 78,000. The data suggests one in 15 searches now draws on social media, reshaping how brands think about visibility in search results.Lire la suite

Nonprofit Current AI unveiled Suno Sutra, a pocket‑sized offline AI device that runs in 22 Indian languages, at the India AI Summit. The organization also announced a $3.2 million grant round for projects in Kenya, Lebanon and Brazil, backed by $400 million in commitments from the French government, the Ford and MacArthur Foundations, DeepMind and Salesforce. Current AI’s CEO Ayah Bdeir said the effort aims to create a public‑interest alternative to privately owned AI systems.Lire la suite

Hugging Face announced that a cyberattack last week breached its internal datasets and exposed service credentials. The intrusion exploited a vulnerability in a user‑uploaded dataset, allowing attackers to run malicious code, elevate privileges, and access internal systems. The company has revoked the stolen keys, patched the flaw, and urged users to rotate any credentials stored on the platform. Law enforcement has been notified, and external forensic specialists are reviewing the incident. Hugging Face said its own AI‑driven anomaly detection flagged the breach, prompting a switch to an in‑house language model for analysis.Lire la suite

Netflix disclosed a $587 million cash purchase of InterPositive, the AI-driven post‑production startup co‑founded by actor‑director Ben Affleck. The deal, announced in March, brings the entire InterPositive team into Netflix and installs Affleck as a senior advisor. InterPositive’s tools aim to fix missing shots, replace backgrounds and correct lighting using generative AI. Bloomberg reported the price could be as high as $600 million. Netflix says roughly 300 of its titles already use generative AI, signaling a deeper push into AI‑enhanced content creation.Lire la suite

Apple filed a trade‑secrets lawsuit against OpenAI, alleging the AI firm recruited current and former Apple staff to obtain confidential information. OpenAI said it sees no merit in the complaint. The case could delay OpenAI’s planned mobile smart speaker and may affect the timing and valuation of its confidentially filed IPO, which analysts expect by year‑end or early next year. Apple named its chief hardware officer, Tang Tan, among those allegedly targeted. The dispute spotlights the growing tension between tech giants over talent and intellectual property.Lire la suite

Author and educator Dave Eggers told roughly 200 OpenAI employees that ChatGPT is crippling teachers and robbing students of their voice. Speaking at a company‑wide session arranged by CEO Sam Altman, Eggers called AI‑generated writing "pastiche nonsense" and warned that reliance on the tool could silence an entire generation of writers. His remarks, reported by the Financial Times, underscore growing concerns among educators about the impact of generative AI on learning and creativity.Lire la suite

Moonshot AI told investors on Saturday it plans to list in Hong Kong within six months, seeking a valuation north of $30 billion. The three‑year‑old startup reported $300 million in annual recurring revenue for June, up from $200 million in April, and unveiled its 2.8‑trillion‑parameter Kimi K3 model, which benchmarked against top U.S. systems. The company is dismantling its offshore VIE structure and has hired CICC and Goldman Sachs to steer the offering. Co‑founder Yang Zhilin, a former Tsinghua professor with stints at Meta and Google, hopes the deal will set a new bar for Chinese AI firms.Lire la suite

Researchers at security firm Tracebit demonstrated that embedding malicious prompts alongside secrets in Amazon Web Services can cripple AI hacking agents. In tests across five leading large‑language models, the technique—dubbed “context bombing”—slashed successful admin‑level breaches from 57 percent to 5 percent and eliminated persistent footholds in most cases, offering a new defensive lever against prompt‑injection exploits.Lire la suite

Chinese AI firm Moonshot unveiled Kimi K3, a 2.8‑trillion‑parameter large language model that the company says matches or exceeds most U.S. offerings except Anthropic’s Claude Fable 5 and OpenAI’s GPT‑5.6. Marketed as the world’s first open‑weight 3‑trillion‑class model, Kimi K3 will be downloadable on July 27 and is priced on par with leading frontier models, while boasting strong coding performance against domestic rival Z.AI.Lire la suite

A new report from Meta's independent Oversight Board reveals that popular large‑language‑model chatbots from six major AI firms consistently refuse or dilute requests that criticize governments in countries with tight speech controls. Testing ten models, the board found higher refusal rates for prompts about China, Thailand and Saudi Arabia than for comparable queries about more permissive nations. The findings raise concerns that AI tools may be acting as a proxy for state censorship, potentially stifling dissent and undermining free expression worldwide.Lire la suite

Apple has lodged a lawsuit against OpenAI, accusing the startup of infringing on its technology and engaging in unfair competition. While the complaint is detailed and aggressive, industry experts say many of the claims mirror standard practices in the fast‑moving AI sector. The legal move comes as Apple rolls out public betas of its new Siri AI, intensifying the rivalry between the two tech giants.Lire la suite

OpenAI announced that parents linked to their teen's ChatGPT account will receive notifications if the account is deactivated for violating the company's policies on violent threats or acts. The move expands the parental‑control suite launched last year, which already lets guardians set usage limits, filter sensitive content, and flag self‑harm concerns. Working with online‑safety firm Moonshot, OpenAI also rolled out a new Study Mode and more frequent break reminders for teens who spend long periods in the chatbot. The changes come after criticism that the firm failed to alert authorities when a mass‑shooter’s account was shut down.Lire la suite

Apple filed a lawsuit Friday accusing OpenAI of stealing confidential hardware designs and prototypes, naming former Apple chief hardware officer Tang Tan among the defendants. The suit arrives as New York Governor Kathy Hochul signs the nation’s first statewide moratorium on large data centers, and health officials warn that a cyclospora parasite outbreak has become the worst on record, with thousands sick across more than 30 states.Lire la suite