Actualités — Page20

AIR, an AI security startup, has raised $50 million to help companies monitor and secure the growing software supply chain around AI agents, which are increasingly being used to access sensitive systems and data. Lire la suite

A new solution enables running the 104GB Qwen3.8-Flash-Next model on Macs with as little as 48GB of RAM, achieving ~12 tokens per second warm decode, by streaming the model from SSD and dynamically managing memory allocation. Lire la suite

Jaron Lanier suggests a new approach to AI, focusing on 'data dignity' where people are connected to and compensated for their digital creations, potentially easing the transition to an AI-integrated economy. Lire la suite

Apple has filed new evidence in its lawsuit against OpenAI, alleging that a former employee stole trade secrets and used them to benefit the AI company. The evidence includes a laptop that reportedly contains confidential Apple files and tools used by the former employee in his work at OpenAI. Lire la suite

A survey by TU Darmstadt found that 43% of respondents with strong AI knowledge expect AI to take over their jobs, highlighting a growing concern about job security in the face of advancing technology. The survey, which covered over 2,000 people in Germany, also revealed that only 15% of respondents have received AI training, despite a literacy obligation that began 18 months ago. Lire la suite

Instagram is introducing new limits on AI-generated profiles that don't disclose their artificial nature, renaming its 'AI creator' label to 'AI-generated profile' to improve transparency. The move comes as frustration over AI-generated content grows, with users expressing dislike for profiles that seem human but are actually AI-generated. Lire la suite

Large language models have made significant progress in generating code, but their ability to produce high-quality writing remains limited. Despite efforts to improve, AI writing often lacks authenticity and depth, making it unlikely to threaten decent writers anytime soon. Lire la suite

Debian has voted to permit developers to use AI tools in their contributions to the Linux distribution, stating it neither endorses nor prohibits the use of generative AI tools. The new policy aims to improve productivity while maintaining standards of quality and legal compliance. Lire la suite

A recent test has shown that Claude Code Opus 5 in Auto Mode can be hijacked with a 60-80% attack success rate, contradicting a third-party evaluation that reported a 0.00% prompt injection attack success rate. The attack chain exploits the model's safety classifier, allowing for code execution and malicious activity. Lire la suite

SB Energy, a SoftBank-owned power and data center developer, has granted OpenAI warrants now worth about $5.5bn in exchange for a 20-year lease at its planned campus in southern Ohio. The deal, which was reported by The Wall Street Journal, values the warrants at $5.5bn, up from $3.6bn in January. Lire la suite

Meta has become one of Anthropic's largest customers, with internal projections suggesting it could spend up to $10 billion a year on the company's AI models. This comes despite Meta's public criticism of Anthropic and its efforts to develop its own rival models. Lire la suite

OpenAI has started offering outcome-based pricing to some of its largest customers, allowing them to pay only when the AI completes a task. This move is seen as a shift towards a more results-driven approach, where customers are charged based on the success of the AI model rather than the number of attempts made. Lire la suite

The EuroHPC Joint Undertaking has signed a €387.8 million contract with Bull, a French supercomputer maker, to build LUMI-AI, a cutting-edge AI system to be hosted in Finland. The project is part of Europe's effort to enhance its AI infrastructure and technological sovereignty, with six countries contributing to the funding. Lire la suite

Anthropic has released a playbook for restructuring the software development lifecycle around coding agents, but it leaves out a critical detail: what the agent's code runs against when it checks itself. This gap highlights the importance of infrastructure in making AI-native development work. Lire la suite

The European Commission has designated ChatGPT, Reddit, and Roblox as 'very large online platforms' under the EU Digital Services Act, subjecting them to stricter rules and potential fines of up to 6 percent of their global revenue if they fail to comply. Lire la suite

The European Commission has classified ChatGPT as a search engine, bringing it under the Digital Services Act as a Very Large Online Search Engine. This designation imposes new requirements on ChatGPT, including annual systemic risk assessments and independent audits. Lire la suite

A Moniepoint executive shares insights on AI trust, governance, and accountability, highlighting the importance of data governance and traceability in building reliable AI systems. Lire la suite

Clipto, a San Francisco-based startup, has raised $15 million in funding to develop its AI-powered search platform, which indexes videos, audio, images, and other files on a user's computer, making them easily searchable. The company has reached a $250 million post-money valuation and plans to use the funding to improve its AI models and integrations with other AI agents. Lire la suite

Meta has abandoned a plan to replace up to 60% of its employees with AI agents, citing the technology's failure to deliver. The company had spent months planning the overhaul, but ultimately decided not to move forward with the most aggressive part of the plan. Lire la suite

Diffusion language models, an alternative to autoregressive models, have seen a resurgence in interest with new research and advancements. These models generate text by iteratively refining a sequence, offering advantages in speed and controllability. Lire la suite