AI Headlines:
November 19, 2025: Agent alliance against tech silos, Meta revolutionizes hearing, Red alert due to “shadow AI”

Just 24 hours after Gartner warned of “AI agent silos” (news from yesterday), the Linux Foundation, Amazon (AWS), and IBM today announced the formation of the “Open Agent Alliance.” The alliance will develop the “Universal Agent Protocol” (UAP), an open standard that will allow AI agents from different vendors to communicate securely and delegate tasks.

Why is this important? Without this standard, SAP’s purchasing agent cannot communicate with the logistics agent in Google Cloud.

The conflict: The alliance is a direct counter-proposal to the closed systems that Microsoft/SAP and Google/Salesforce are currently building. Experts see this as the beginning of the next major “platform war”: open standards versus proprietary ecosystems.

Meta today rolled out a massive software update for its latest generation of smart glasses, revolutionizing audio processing. The new “Audio Focus” feature uses AI to artificially create the so-called “cocktail party effect.” Wearers of the glasses can look at a single person in noisy environments (e.g., exhibition halls or restaurants), and the AI ​​will isolate and amplify their voice while almost completely blocking out all ambient noise.

The enhancement: Simultaneous translation has also been improved. The glasses now not only translate but also synthesize the translated language directly into the speaker’s voice and deliver it to the wearer’s ear. This is being hailed as a breakthrough for international business travel and accessibility.

Security provider CyberArk today released its “Global Threat Report 2025,” sounding the alarm. Despite strict prohibitions and the introduction of official tools (such as Microsoft Copilot), 62% of surveyed employees continue to secretly use private, unauthorized AI tools (“shadow AI”) for their work.

Employees often perceive the official, security-restricted corporate AI as “too dumb” or too slow compared to the latest models from startups or open-source solutions.

The risk: The report documents cases in which highly sensitive research data and internal financial reports were uploaded into the training data of public “cheap” AIs. This indirectly confirms Apple’s strategy of “clean on-device AI,” which was presented yesterday as a counter-proposal.

November 19th illustrates the central tension of the current phase: On the one hand, the attempt to bring order to the chaos of the systems through standards (Story 1). On the other hand, the “uncontrolled growth” caused by users who do not adhere to the rules (Story 3) because the technology (Story 2) is simply too tempting and useful.