PromptArmor, an AI security firm, reported that the number of tools available to connectors used by OpenAI's ChatGPT and Anthropic's Claude has increased significantly over a six‑week period, expanding the potential risk surface for AI agents. Connectors allow AI agents to interact with third‑party services such as Gmail or Slack, and their rapid evolution complicates security assessments. From mid‑May to the end of June, 37 percent of the 2,517 connectors examined were updated, 1,686 new tools were added, and 1,127 tool descriptions were rewritten. The Dropbox connector grew from eight to 24 tools, including ten write‑capable functions and four potentially destructive ones. An evaluation of 487 Claude connectors found that about 2 in 5 are likely to invoke additional AI services, which may process data on external infrastructure. A case involving the Zoom connector showed how sensitive queries could be transmitted to multiple third‑party AI models, raising data‑privacy concerns. Connector security controls provided by the AI developers do not govern the processing performed by third‑party services, which may occur outside the United States. These findings indicate that ongoing monitoring of connector toolsets is necessary to manage emerging security risks associated with AI agents.