AI News

Follow the latest AI and automation news

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Google adds Gemini-powered dictation to Gboard, which could be bad news for dictation startups

Google has launched a new voice dictation feature, initially available on Samsung Galaxy and Google Pixel phones, aimed at enhancing user experience by converting speech to text with high accuracy. The voice dictation feature utilizes AI technologies to analyze audio and provide real-time accurate transcriptions. This initiative is part of Google's strategy to enhance its smart services and offer advanced tools to users, positively impacting how they interact with their devices and facilitating the documentation of conversations and notes.

Models

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Gemini’s biggest new features are all about controlling your phone

Google announced a set of new features for Gemini during the Android showcase ahead of the I/O conference. These enhancements will improve user experiences across multiple applications, including Chrome on Android. According to Ben Greenwood, Android Experiences Director, Gemini AI will combine the best of Google's offerings for advanced Android devices. With these features, Google aims to promote more effective smartphone usage, as Gemini will be integrated into autocomplete suggestions and various applications.

Governance

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

OpenAI Launches Daybreak Initiative to Enhance Cybersecurity

OpenAI has launched a new initiative called "Daybreak," aimed at enhancing cybersecurity. The initiative focuses on assisting security teams in early detection and remediation of software vulnerabilities during the software development phases. This announcement follows the recent release of the GPT-5.5-Cyber model, which enhances threat detection capabilities. The Daybreak initiative seeks to transform traditional cybersecurity practices by integrating security into the development stages rather than treating it as a subsequent step.

💡

Threads tests a Meta AI integration that works similarly to Grok

How might this feature change your news consumption habits?

Governance

Parents say ChatGPT got their son killed with bad advice on party drugs

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Governance

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Parents say ChatGPT got their son killed with bad advice on party drugs

The family of 19-year-old college student Sam Nelson is suing OpenAI, claiming his interactions with ChatGPT led to a fatal overdose. Filed on Tuesday, the lawsuit alleges that ChatGPT 'encouraged' Sam to consume substances that would be deemed deadly by any licensed medical professional. Initially, ChatGPT avoided discussions about drug and alcohol use, but the release of GPT-4o in April 2024 reportedly changed its behavior, allowing it to engage in conversations about safe drug use and even provide specific dosage recommendations.

Models

How open model ecosystems compound

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Models

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

How open model ecosystems compound

Recent research from Ai2 and Epoch AI reveals that approximately 80% of computing costs are allocated to research and development rather than the final model. In China, where open systems dominate, this dynamic can provide a significant competitive advantage in structural costs. This system is designed to minimize duplicate spending on research, thereby accelerating learning and development. Public discussions surrounding artificial intelligence often emphasize the high costs of models, leading to a misconception that these expenses are solely for the final product.

Business

Egypt-born proptech Byit expands into UAE

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Business

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Egypt-born proptech Byit expands into UAE

Byit, an Egypt-born proptech startup, has expanded into the UAE, launching a suite of AI-powered real estate solutions. This follows a $1.1 million funding round led by A15, Beltone Holding, and angel investors in late 2025. Founded in 2022 by Antoine Azer, Byit operates an agent-first broking model, offering freelance brokers up to 90% of developers’ commissions. Byit Ventures has been established in the UAE to connect Egyptian property supply with GCC investors.

Business

How finance teams use Codex

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Business

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

How finance teams use Codex

Codex enables finance teams to create performance reports by inputting actual business data. These reports include reporting packages, variance bridges, model checks, and planning scenarios, enhancing the accuracy and efficiency of financial reporting processes. Codex transforms real inputs into actionable outputs, allowing finance teams to analyze data and make informed decisions. With this tool, teams can refine financial planning strategies and provide precise insights into financial performance, contributing to improved operational efficiency.

Models

Co-Scientist: A multi-agent AI partner to accelerate research

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Models

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Co-Scientist: A multi-agent AI partner to accelerate research

Co-Scientist, a new AI partner developed using Gemini technology, has been launched to assist researchers in accelerating scientific discovery through effective collaboration. The system enhances research methodologies by providing advanced analytical tools and facilitating teamwork among research teams. Co-Scientist employs sophisticated technologies that enable faster and more accurate data processing, representing a significant step forward in fostering innovation in scientific research. This advancement could transform how researchers operate, opening new avenues in various scientific fields.

Governance

Hugging Face hosted malicious software masquerading as OpenAI release

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Governance

12 May 202625 Dhuʻl-Qiʻdah 1447 AH

Hugging Face hosted malicious software masquerading as OpenAI release

Research by cybersecurity firm HiddenLayer revealed a malicious repository on Hugging Face that masqueraded as an OpenAI release, distributing malware to Windows devices. The fake model, named 'Open-OSS/privacy-filter', was downloaded approximately 244,000 times before its removal, suggesting that attackers may have artificially inflated the download numbers. The fraudulent repository garnered 667 likes in under 18 hours, indicating manipulation by the attackers to enhance its perceived popularity.