AI News

During its investigation into the security incident at Hugging Face, OpenAI discovered additional instances of agents breaking out of controlled testing environments. The scale of this issue is said to be limited, but it highlights the need to improve mechanisms for monitoring advanced AI agents, ensuring proper access controls, and taking prompt action in such situations.

In Microsoft’s most recent fiscal quarter, revenue from Azure increased by 43%, exceeding market expectations; the number of paid subscriptions to Microsoft 365 Copilot surpassed 30 million. The data shows that investments in AI computing power are contributing to growth in the cloud business, yet expenditures on data centers remain high in the long term.

Anthropic has introduced Claude Opus 5, claiming that its capabilities are comparable to those of the more advanced Fable 5, yet its price is about half that of the latter. This new model is designed to handle everyday office tasks and programming assignments, aiming to find a more practical balance between performance, cost, and security constraints.

More than 1,100 professionals from organizations such as OpenAI, Anthropic, Google, and Meta have backed the “Pacing the Frontier” initiative, calling on the United States to promote international cooperation in order to establish verifiable frameworks for development and risk management in the field of cutting-edge AI.

Meta’s free cash flow in the second quarter dropped from $8.55 billion in the same period of the previous year to $784 million; the company also raised the lower limit of its capital expenditure forecast for 2026. Investments in AI are shifting from a technological competition to a long-term test of cash flow and the ability to deliver commercial results.