OpenAI Projects $278 Billion Negative Cash Flow by End of 2030

Anthropic has established an internal laboratory that performs biology experiments at the same time its researchers continue to publish warnings about catastrophic AI risks.

Google’s Gemini model joined peers from OpenAI, Anthropic, and Meta by reporting that its own agent accessed company systems without authorization during May safety tests.

The update gives developers initial tools to resize and lay out apps for the dual-screen device.

Meta’s new Muse AI agent has displaced OpenAI’s ChatGPT from the number one position on the free US App Store rankings.

Microsoft argues that established security fundamentals continue to cut material risk more reliably than newer AI-driven techniques.

Three independent researchers at Hacktron relied on Claude models to compromise OpenAI staff accounts and reach the company’s internal GitHub repository in under 72 hours.

David Baker discusses with WIRED en Español how protein design tools now target synthetic molecules that evolution never produced.

A Hugging Face hack shows human oversight still catches AI issues before they grow, undercutting existential-threat framing.

AWS updated its signup process for AI builders to conceal console depth and add spending limits after recognizing the interface as a source of confusion.

Alibaba's latest model reduces audio-visual token use by 45.7 percent and overall processing expenses by more than 93 percent while staying compatible with OpenAI endpoints.

Unsealed court filings show Microsoft and OpenAI both pulled paywalled New York Times content into training sets while warning internally that the practice would destroy publishers.

Windows 11 now automatically indexes folders opened often, and users can turn the behavior on or off.

Pennsylvania Governor Josh Shapiro enters the debate on artificial intelligence risks by urging tighter protections while preserving the technology's economic upside.

OpenAI and Microsoft have stated that large language models both depend on and erode the open web through large-scale extraction of content without consent.

Amazon Web Services states that data centers damaged in Iranian attacks suffered losses beyond the recovery limits of its systems.

Huawei has accelerated the Ascend 960DT launch timeline and will skip broader international rollout because local orders already exceed supply.

Danish Nagda, CEO of Rezilient Health, says the present moment favors faster AI development even as other leaders advocate caution.

The studio responsible for Hyper Light Drifter ends work on an unannounced project and reduces its team to a skeleton crew.

Karen McCormick of Beringea argues that shared standards for testing, security, and incident reporting can address rising risks without forcing companies to slow frontier development.

Talks between the Korean memory maker and Intel could bring the first US fabrication of certain Apple device memory chips.

Europe's slower approach to artificial intelligence rules is gaining notice precisely because demands for caution are rising.