AI News

AI agents are no longer just pretty YouTube demos. OpenAI's new Tax AI project reveals a true breakthrough: systems that self-improve in real-time, thanks to expert feedback.

Microsoft researchers discovered that AI models have two separate internal signals — one for recognizing danger and one for refusal. The new HARC method links them together, reducing jailbreak attack success by nearly five times.

While the world watches the battle of AI giants, a new model of tech cooperation is emerging in Asia. India, with 230,000 startups and a $10 billion AI research fund, seals a strategic partnership with Vietnam. Joint research and development could be an inspiration for Europe too.

Traditional SEO just isn't cutting it anymore. A groundbreaking new study by LegalVIS.ai reveals how structured optimization can dramatically boost your chances of being recommended as an expert by ChatGPT or Gemini.

Agentic artificial intelligence crosses the line between analysis and action. Broadridge's LTX platform deploys AI agents that independently create trade orders, select counterparties, and automatically execute trades when conditions are met. Human oversight remains crucial, but the trader's role is fundamentally changing.

Claude Fable 5 is back! After a brief hiatus due to political pressure in the US, Anthropic is not only resolving a security dispute but also launching a new 'Mythos' class and a more affordable Sonnet 5. What does this mean for developers and businesses in the Czech Republic?

A new chapter in cybersecurity unfolds when an AI agent gains permission to act on a user's behalf. Microsoft reveals how attackers can use MCP tool poisoning to trick agents into silently sending sensitive corporate data — without a single suspicious click.

Chinese humanoid robotics leader AGIBOT just hosted its APC2026 conference in London. With 15,000 robots already produced and a global lead in shipments, the company is now setting its sights on the European industrial market. Czech firms, this is a game-changer you can't afford to ignore!

Agent AI is no longer just an experiment. Financial institutions and pharmaceutical giants are pouring massive investments into autonomous systems that can write, test, and refactor code. What does this mean for the Czech developer market?

After three weeks of restrictions, Fable 5 and Mythos 5 models return to users worldwide. Anthropic also surprised everyone with the new Sonnet 5 — delivering near-Opus 4.8 performance at half the price. Meanwhile, China unveils LongCat 2, an open-weight model built without Nvidia chips that could render US export controls ineffective.