AI is changing what is worth knowing
Most of the EU AI Act that was due to apply on 2 August was moved to 2027. The requirement that staff be AI literate stayed, but was rewritten from ensuring a level to supporting the development of one.
Fable 5 took 6 percent of the tokens companies bought from Anthropic. Anthropic's own cheaper Opus 5 passed it within a month. The spending never left Anthropic. It moved further down the company's own catalogue.
Anthropic has launched Claude Tag, moving the AI assistant into Slack as a shared actor anyone in a channel can tag and delegate tasks to. The real shift is not the platform but the way of working, from asking a tool to delegating to a colleague with memory, permissions and traceable assignments. For Swedish organisations the question is what happens when an AI gets access to the place where work is already coordinated, which makes permission design the control point.
OpenAI has launched Patch the Planet, an initiative using AI models to find security flaws in critical open source software. The results show that AI can already detect advanced vulnerabilities in systems like OpenBSD, Chrome, Safari and Firefox, but also that every finding still needs human review. The big question is therefore not only how fast AI can find problems, but whether the open source world has the capacity to handle the new pace.
Anthropic is expanding Project Glasswing to around 150 more organisations in critical infrastructure across more than 15 countries. Sweden is on the list. The real problem is no longer finding the vulnerabilities, it is fixing them in time.
The European Commission and OECD's new AI Literacy Framework makes AI literacy more than a question of which tools are allowed in the classroom. For Swedish schools it points at a governance question that authorities and head teachers cannot delegate away.