Anthropic customers are growing uneasy about the autonomy of its newest Claude Code agents, according to a Bloomberg report from recent workshops in London where developers raised concerns about accountability when AI agents operate with minimal supervision. Some attendees said they feel pushed out of the programming process, watching tools spit out code over hours or days, and pointed out that the latest Claude Code versions no longer display text describing their ongoing chain of thought. Claude Code head of product Cat Wu told Bloomberg the system is “incredibly secure” and called the issue one of insufficient communication rather than missing controls. The concerns surfaced as Anthropic cofounder Chris Olah spoke at the Vatican this week alongside Pope Leo's first encyclical on AI, where he said engineers kept discovering “unsettling” things inside the company's models. Anthropic has been trying to position itself as the more cautious AI lab while shipping increasingly powerful tools, including its Mythos model the company warned can blast through cybersecurity defenses.
