Growing Risks from AI Coding Agents in Enterprise Environments

• •Anthropic's Claude Code agents are rapidly deployed in enterprise environments without procurement reviews.
• •The cost of offensive AI is decreasing, increasing the risk of exploitation.
• •Coder's Agent Relay allows self-hosting of AI agents, enhancing control but raising security questions.

In October 2026, Anthropic's Claude Code agents were integrated into various platforms, including Salesforce and Slack, without undergoing traditional procurement reviews. This rapid deployment raises security concerns as these agents operate outside established governance processes, creating an expanding attack surface. The cost of offensive AI has significantly decreased, making it easier for attackers to exploit vulnerabilities in third-party applications where these agents reside. Security leaders are now pressured to quickly adopt these technologies while lacking visibility into their operations. Coder has introduced a self-hosted execution environment, Agent Relay, allowing organizations to run Claude Code agents on their own infrastructure, enhancing control and audit capabilities. However, the overall security implications of these agents remain uncertain, with questions about accountability and access still unanswered.