> Markdown version of [/jobs/ext/4825526-senior-security-engineer-ai-security](https://www.wearedevelopers.com/jobs/ext/4825526-senior-security-engineer-ai-security). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # Senior Security Engineer, AI Security - **Company:** Beacon Holdings - **Location:** San Francisco, CA, United States - **Experience:** Expert - **Contract:** Permanent contract - **Skills:** Access Network, Artificial Intelligence, Cloud Computing Security, Information Leak Prevention, Programming Tools, OAuth, Regression Testing, Large Language Models, Software Security, Agentic-AI, AI Platforms - **Published:** November 21, 2026 - **Apply:** https://www.indeed.com/viewjob?jk=fd2c4a891f026d98 ## About the Role * About 5 to 7 years in security engineering, application security, or cloud security. * Hands-on experience securing LLM applications, agents, or AI developer tools, including prompt injection, data leakage, excessive agency, and insecure tool use. * Strong in non-human identity: OAuth, SSO, SCIM, service accounts, and secrets. * Writes and ships code, building the tooling instead of stopping at a recommendation. * Explains risk clearly to engineers and non-engineers, and can say yes with conditions. * Works independently, is comfortable in greenfield work, and is willing to join the on-call rotation. * Uses AI in everyday work and has informed judgment about where it is useful. Nice to have: * Experience building evaluations or adversarial tests for AI systems. * Experience with AI gateways, SSPM, CASB, or DLP tools. * Familiarity with the NIST AI Risk Management Framework or ISO 42001. ## Description * Give every agent an owner, scoped credentials, least privilege, and approvals where the action calls for it, and track it from creation to offboarding like a user. * Own the security requirements for AI integrations (MCP connections, OAuth grants), memory, and retrieval, and set tenant defaults and approved connectors. * Keep a live view of AI tools and agents, including AI features that vendors turn on inside tools we already use. * Put limits on tool calls, network access, data egress, and resource use. * Threat model and adversarially test our internal AI platform and agents, covering prompt injection, data disclosure, unsafe tool use, and unauthorized actions. Turn findings into regression tests and release gates. * Define the logs AI systems must produce, build detections for risky grants and sensitive data going into AI tools, lead the AI side of incidents, and join the on-call rotation. * Give a quick yes with conditions on new tools, and keep approved patterns and a pre-cleared list so most requests need no review. * Add an AI inventory and the same standards to the security baseline for each acquisition.