Exaforce adds a kill switch for AI agents that go rogue
Agentic security operations startup Exaforce Inc. today launched Exaforce AI Security, capabilities that let security teams see the artificial intelligence agents running across their environments and shut down the ones that turn hostile. Exaforce built the release around a gap in how enterprise logs record what agents do. Agents act with the identities and permissions […] The post Exaforce adds…
Startup Exaforce Inc. has unveiled Exaforce AI Security, a new solution that enables security teams to monitor artificial intelligence agents operating within their systems and terminate those that exhibit malicious behavior. This development addresses a significant gap in enterprise log records, which typically only reflect agents' actions through the identities and permissions of the individuals deploying them.
When an agent performs actions like rotating keys or deploying code, the audit trail typically appears benign. Attackers have exploited this weakness, as evidenced by the June 2025 compromise of Klue Labs Inc. and the August 2025 breach affecting over 700 organizations through Salesloft Inc.'s Drift chatbot. The new Exaforce AI Security offers four key features.
Initial agentless discovery runs continuously, identifying AI applications, coding agents like Claude Code and Cursor, hosted agents, custom GPTs, Model Context Protocol servers, skills, and development environment plugins, all linked to their respective users and permissions. A risk layer analyzes these findings, highlighting excessive permissions or unauthorized applications.
Threat detection and automated response mechanisms complete the suite. Exaforce's platform cross-references the activity of AI agents and models with human identities, endpoint telemetry, file access, and code context, uncovering misuse and instances of sensitive data exposure that would remain unnoticed through individual log entries.
Upon detecting anomalies, the platform can disable a session, deactivate a model provider key, isolate a device, or terminate an agent's process, referred to as an agent kill switch. Security teams can adjust the autonomy of each action, ranging from analyst-approved to fully automatic. The platform aggregates endpoint data, productivity suite activity, and model provider audit and usage logs, feeding them into Exaforce's knowledge graph, which already integrates identity, cloud, software-as-a-service, and code sources.
The release expands an existing integration with Anthropic PBC's Claude Compliance API, adding support for OpenAI Group PBC's ChatGPT, Google LLC's Gemini, and Microsoft Corp.'s Copilot. Exaforce AI Security is now generally available on the company's agentic security operations platform, available for self-deployment or through its managed detection and response service.
Funded by investors including HarbourVest Partners, Peak XV Partners, Mayfield Fund, Khosla Ventures, and Seligman Ventures, Exaforce raised a $125 million Series B round in May, bringing its total funding to $200 million since its inception in 2023.
Written by urgent.news from SiliconANGLE's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
