Anthropic has released a new version of its AI assistant that lives directly in Slack - the corporate messaging tool used by millions of companies. It's called Claude Tag and is designed as „a team member that never sleeps“: you tag it in a channel, it gives you back analyses, takes on tasks and - this is the key part - constantly reads and remembers what the team writes.

The idea sounds practical. Instead of searching old messages, you ask the assistant. It breaks tasks into steps, replies in threads with finished work, and in the so-called „ambient mode“ it flags problems on its own. Administrators decide which channels it can access - the legal Claude doesn't look into the engineering one, and vice versa. On paper, everything is under control.

The question that polished press releases rarely raise is a different one: what does it mean when a company gives an artificial intelligence constant access to every work message? Every joke, every grievance, every mistake spoken in confidence among colleagues becomes material the model learns from. The tool isn't malicious - but the line between „an assistant that helps“ and „a system that records everything“ is thinner than the marketing wants to admit.

Anthropic isn't alone in this race. The same market sees Microsoft with its Copilot, as well as Glean, Snowflake and Databricks - all promising to „understand“ what your company does. The winner will be whoever convinces the most companies to open the doors of their own conversations. The employee writing in those channels is rarely asked.