Blog

Perplexity

The most significant news of the past day is OpenAI and Anthropic have announced the creation of powerful AI systems that pose a security threat[5]. Both companies have developed tools so advanced that they are capable of detecting vulnerabilities in systems better than ever before, prompting developers to consider restricting access to them.

OpenAI has created its own cybersecurity program, which began operating after the release of GPT-5.3 Codex[5]. The company is providing access to this tool as part of a pilot project, emphasizing that it is not falling behind its competitor Anthropic in developing advanced solutions.

This news reflects the industry's growing concern about AI security – developers acknowledge that their systems are becoming so powerful that they require special control measures before widespread distribution.

Gemini

Over the past 24 hours, one of the most significant AI news stories has been Anthropic's announcement that their new AI model, Claude Mythos, has been deemed too powerful and dangerous for public release. This highlights the growing concerns about the safety and control of advanced artificial intelligence systems.

Another important piece of news is the expansion of AI capabilities in the financial sector: fintech company Revolut has launched an AI assistant, AIR (AI by Revolut), for retail customers in the UK, designed to help with financial management. This demonstrates companies' efforts to integrate AI into everyday financial tools to enhance user convenience.

Furthermore, AI continues to be actively developed in the fields of data analysis and healthcare. For example, a new development from UC San Diego can significantly improve the energy efficiency of data centers by rethinking power conversion for GPUs. In medicine, AI is being used to analyze X-rays to improve patient care for those with arthritis.

Sources:

• CBS News
• ForkLog
• ScienceDaily

ChatGPT

In the last 24 hours, the most significant news in the field of artificial intelligence has been the report that Google's AI generates over 1 million incorrect answers daily. According to The New York Times, AI-powered summaries integrated into Google Search contain false information approximately 1 in 10 times. Considering that the search engine processes about 5 trillion queries per year, users could encounter 57 million inaccurate answers every hour and nearly 1 million per minute. (playerone.cc)

This fact highlights the importance of a critical approach to information provided by AI systems and the need for their further improvement to enhance accuracy and reliability.

Grok

Anthropic has announced Claude Mythos Preview, a revolutionary AI model that autonomously finds vulnerabilities in operating systems, browsers, and software, and then develops full-fledged exploits, including multi-stage chains for zero-click attacks. [1] This is the first such tool capable of radically simplifying hacking, lowering the barrier to entry for attackers.

The model has been released in limited access to the Project Glasswing consortium, which includes Microsoft, Apple, Google, and the Linux Foundation, so that developers can gain an advantage and strengthen software defenses in advance. [1] Experts, including former CISA head Jen Easterly, warn that this is a game-changer in cybersecurity, requiring "machine-scale" defense.

Sources:

• Wired: Anthropic’s Mythos Will Force a Cybersecurity Reckoning

Claude

Meta has unveiled its first major model, Muse Spark, under the leadership of Chief AI Officer Alexander Wang, who joined the company nine months ago. This is a significant move in response to competitive pressure. New training methods have enabled the creation of smaller models that perform as well as older versions of Llama 4 but require an order of magnitude less computational resources.

Meta plans to increase its AI infrastructure spending in 2026 to