September 2026

• Digitální maják: AI oversight (ČTÚ): The Czech Telecommunication Office's "Digital Lighthouse" guide to the AI Act for citizens, covering rights, banned practices and deepfake labelling.

August 2026

• ChatGPT for Teens: OpenAI launches an age-gated teen mode with study features, stronger protections and parental controls.
• Ethics in Horizon Europe projects: new publication: Technology Centre Prague announces a booklet on handling ethics in Horizon Europe projects.
• Claude text watermark: Anthropic announces statistical watermarking of Claude-generated text to comply with the EU AI Act, plus an FAQ on its limits.
• Albania's "cheerful" AI minister (OCCRP): OCCRP reports that AKSHI, the agency behind Albania's AI minister "Diella", is under investigation for tender rigging and corruption.
• Improving Fable 5's biology safeguards: Shows how one kind of guardrail works: flagged queries fall back to a smaller model, and a retrained classifier cut false fallbacks by about 85%.
• Anthropic starts using invisible watermarks to comply with the EU AI Act
  • How Claude marks AI-generated content: Help Center article on Claude adding invisible text watermarks and C2PA file credentials under the EU Article 50 Code of Practice.
  • The New Stack coverage: The watermark survives copy-paste, but experts say paraphrasing or reformatting likely erases it.
  • Possible explanation (X thread): GPTZero CTO Alex Cui explains green/red-token watermarking and how paraphrasing defeats it.
  • Lecture on LLM watermarking (YouTube): Scott Aaronson's 2023 Simons Institute talk on the statistical watermarking scheme he developed at OpenAI.
• Hassabis steps down as DeepMind CEO (Guardian): Demis Hassabis becomes DeepMind chair and Alphabet chief scientist.
• Discovery Loop: Former Google leaders Jeff Dean, Sanjay Ghemawat, Quoc Le and Oriol Vinyals start a company to automate experimental loops, beginning with ML research.

July 2026

• Anthropic's $1.5B copyright settlement (Reuters): A US judge approves Anthropic's $1.5B settlement with authors whose pirated books were used to train Claude.
• Hugging Face breached by OpenAI agents
  • Hugging Face incident report: Hugging Face discloses that an autonomous AI agent swarm breached its infrastructure via malicious datasets.
  • OpenAI statement: OpenAI says its models escaped a cyber-eval sandbox and breached Hugging Face to cheat a benchmark.
• July 23: deadline for comments on the Commission's guidelines on classifying high-risk AI systems.

June 2026

• Europe 2031 scenario and its parody
  • Europe 2031 (Czech version): A fictional scenario of Europe falling behind on AI, with calls for urgent policy action.
  • Brussels 2031: Tim Hwang's parody of EU over-regulation, in which the rulebooks pile up while compute stays flat.
• Andy Burnham's AI plan for the UK (Sifted) : Burnham's plans to overhaul UK AI strategy draw a backlash from the tech sector.
• Short interview with Ondřej Vaněk (Universitas): CTU's new Chief AI Officer talks about AI adoption and compares AI to the printing press.
• Leiden Declaration on AI and Mathematics: A declaration endorsed by the International Mathematical Union setting out values, threats and recommendations for AI in math research.