Actualités — 2026-06-23

Google’s DeepMind unit is pouring $75 million into A24’s newly formed Labs team, a 20‑person group tasked with creating artificial‑intelligence tools for film production. The partnership gives A24’s filmmakers access to AI‑driven storyboarding and other creative applications while feeding the technology back into Google’s broader AI ecosystem. A24 Labs, led by former Adobe executive Scott Belsky, says the aim is to preserve artistic control, not to cut costs. The deal marks DeepMind’s first equity stake in a Hollywood studio and expands its research beyond video games into cinema. Lire la suite

Anthropic announced a privacy‑policy update that will require a small subset of Claude users flagged for possible policy violations to upload government‑issued identification and a selfie or video for identity verification. Effective July 8, the change introduces the collection of facial geometry templates, a biometric data type covered by state privacy laws such as Illinois' Biometric Information Privacy Act. The verification process is handled by Persona, a San Francisco‑based identity‑checking platform backed by Founders Fund. Anthropic says the measure offers a path to appeal for users rather than an outright ban. Lire la suite

Anthropic announced that its upcoming privacy policy, effective July 8, will ask a limited group of Claude users to verify age and identity by uploading government-issued IDs and a selfie. The move, aimed at letting flagged accounts appeal suspensions rather than face outright bans, comes as the AI firm seeks to satisfy mounting regulatory demands and pressure from the Trump administration. The company will partner with verification service Persona, which will store the documents temporarily, though details on data deletion remain unclear. Lire la suite

OpenAI announced Monday a new program called “Patch the Planet” that will pair its security tools with the expertise of Trail of Bits to help open‑source maintainers fix vulnerabilities faster. The initiative aims to reduce the flood of security reports that small project teams must triage, by having Trail of Bits engineers review findings, develop patches and tests, and embed reusable workflows. OpenAI will supply its Codex Security technology to support the effort. The collaboration targets the broader problem of insecure open‑source code that can ripple into commercial software. Lire la suite

At Meta's Scale conference on Friday, Anthropic co‑founder Boris Cherny answered a heated question about "loops" with a decisive "yes, they're for real." Cherny outlined how AI agents are moving from writing code to prompting other agents that then generate code, and how continuous, self‑reinforcing loops could reshape software development. He described two internal loops—one optimizing architecture, another unifying duplicated abstractions—that submit pull requests autonomously. While praising the potential productivity gains, Cherny warned that the token‑intensive nature of perpetual loops could drive up costs for many firms. Lire la suite