Blog

AI Digest for the Week

In the field of artificial intelligence, this week saw a series of notable events reflecting both technological breakthroughs and the challenges facing the industry[1]. Alisa AI creates jewelry based on text descriptions[1]. Russia is actively discussing the regulation of the AI sector: a meeting was held at the Government's Coordination Center under the leadership of Deputy Chairman Dmitry Grigorenko[1]. Participants are working out key aspects of regulation: quality and safety requirements for AI systems, rules for using data for training, issues of liability for algorithm errors, and the balance between domestic and foreign models[1]. The most important tasks became the legislative consolidation of the concept of "artificial intelligence," defining the boundaries of responsibility for developers and users, and forming a list of areas where the application of AI is permissible or restricted[1]. Consolidated proposals for forming state policy in this area will be prepared following the meeting[1].

The company OpenAI announced the GPT-5.2 model — the most advanced AI system to date, improving overall "intelligence," programming skills, and work with long contexts[1]. Yandex significantly expanded the functionality of its smart cameras and taught them to "report", enabling them not just to record motion but to meaningfully describe what is happening[1]. The device is no longer limited to faceless notifications but reports, for example, that children have returned from school or a courier has left a package[1].

Over the past few days, several major players have released product updates ranging from new chatbot capabilities to image generation systems and integrations between design and programming[2]. Rapid development of artificial intelligence (AI) continues to change the technology market — companies are updating language models, introducing new tools for developers, and testing new ways of regulating AI content[2]. OpenAI presented an update to one of the most widespread versions of the ChatGPT chatbot — the GPT-5.3 Instant model[2]. Anthropic began implementing a voice mode in the developer tool Claude Code[2]. Google announced the launch of an updated image generation model, Nano Banana 2, which became part of the Gemini ecosystem[2]. The design platform Figma announced a partnership with OpenAI and the integration of the OpenAI Codex programming tool[2].

The week was busy: Chinese models are catching up with leaders in reasoning and agency, OpenAI lowered the entry threshold for Codex, and Google is simultaneously testing space data centers and launching its node builders to compete with n8n and make.com[3]. Generative neural networks include a teaser for Nano Banana 2 (4K video), Fish Audio S1 (a model for voice dubbing and cloning), and Scribe v2 Realtime (real-time speech transcription)[3]. AI tools and platforms include Google Opal (a node interface) and free presentations in Gemini[3]. AI in society and research includes Google's space data centers on TPU by 2027, free access to ChatGPT Edu for teachers in Kazakhstan, Sam Altman's suggestion that AI could become OpenAI's CEO, AI writing more than half of new internet articles, and neural networks taking a third of creative jobs in just two years[3].

Microsoft announced massive layoffs, explaining this step by the accelerated introduction of artificial intelligence into key business processes[4]. Google continues to integrate artificial intelligence into users' daily lives: the AI assistant Gemini is now available directly in popular messengers WhatsApp, "Messages," and "Phone"[4]. Meta announced the creation of a new division to develop next-generation AI[4]. Apple did not stay aside: at the WWDC 2025 conference, the company presented new Apple Intelligence features for its entire device ecosystem[4]. Developers gained access to the Foundation Models Framework, opening new horizons for creating intelligent applications[4].

Yandex opened free access to the most powerful version of its neural network Alisa, removing all subscription restrictions[4]. Huawei opened access to its Pangu language models, providing developers with tools to create their own AI solutions even under sanctions[4]. At the regulatory level, the European Union confirmed its intention to implement the AI Act within the established timeframe, despite protests from over a hundred technology companies[4]. Cloudflare is also strengthening control over data use: AI crawlers are now blocked by default for new sites, and resource owners can monetize access to their content for training neural networks[4]. In medicine, a real breakthrough occurred: Microsoft presented the MAI-DxO diagnostic system, which is four times more effective than doctors in diagnosing complex conditions[4].

OpenAI presented the GPT-5.6 model family[6]. OpenAI opened limited access to GPT-5.6 Sol, Terra, and Luna via API and Codex for a small group of trusted partners[6]. The Pentagon expanded the role of AI in military strategy, secretly approving an updated doctrine for using AI to select targets on the battlefield[6]. OpenAI launched GPT-5.5-Cyber amid Anthropic restrictions, a specialized model for finding, verifying, and fixing vulnerabilities[6]. Five Eyes warned about the acceleration of AI cyberattacks, stating that advanced AI models could change offensive and defensive capabilities in cyberspace within months rather than years[6]. Researchers from Nvidia, Carnegie Mellon University, and UC Berkeley presented ENPIRE, a framework allowing AI programming agents to improve robot control policies on real hardware[6]. The "Genesis Mission" AI initiative of Donald Trump's administration includes tasks related to the US nuclear weapons complex[6].

US authorities partially lifted the ban on using the Anthropic AI model[7]. The White House asked OpenAI to delay the release of a new AI model[7]. NYT learned about OpenAI's plans to move its IPO to 2027[7]. The US developed an aircraft for the nuclear arsenal using AI[7]. Anthropic accused Chinese Alibaba of illegally using the AI Claude[7]. AP: Anthropic AI found vulnerabilities in the US government's secret systems[7]. Elon Musk lost his status as a trillionaire[7]. Oracle cut 21,000 employees due to AI implementation[7]. The intelligence alliance of five countries warned about the threat from the latest AI models[7].

AI agents in Telegram, Cursor in SpaceX, and OpenAI in crisis are among the latest neural network news[5]. ChatGPT in Excel, plugins in Claude, and a chatbot in YouTube are among the latest AI news[5]. Over the past few days, major AI players have synchronously taken steps toward integrating neural networks into familiar work tools[5]. Toyota is strengthening the race between Japan and China in robotics[10]. Claude Design is going head-on against Figma, Adobe, and Canva[10]. A robot ran a half-marathon[10].

Perplexity

1. June 23, 2026: OpenAI released a new generation of generative models GPT-5 with a simplified hierarchy (standard, Pro, mini, nano), available to all ChatGPT users, and also opened gpt-oss to the network — the first open models with 120b and 20b parameters for local use[2].

2. June 24, 2026: Google DeepMind presented Genie 3 — a generative AI capable of creating interactive 3D worlds from text descriptions, marking the company’s first open-model release in six years[2].

3. June 25, 2026: OpenAI and Broadcom signed a deal worth more than