Cybersecurity startup Lasso launched its AI Security Posture Management (AI-SPM) platform on July 27, a system designed to solve a problem most enterprise security teams report as their top gap: knowing what agents exist in their organization and what those agents can do.
The platform, detailed in a Hacker News technical overview, operates across four stages: discovery, posture management, automated red teaming, and runtime enforcement. Each stage feeds the next in what Lasso calls a “closed defensive and offensive loop.”
Discovery: Building an AI Bill of Materials
Lasso connects to three categories of agent sources. It integrates natively with managed platforms where business teams build agents without code: AWS Bedrock, Microsoft Azure AI Foundry, Microsoft Copilot Studio, Google Vertex AI, Salesforce Agentforce, and Anthropic Managed Agents. It scans cloud environments (AWS, Google Cloud, Azure) for agents assembled directly in infrastructure without a managed AI service. And it monitors CI/CD pipelines and code repositories for homegrown agents at the build stage, before they reach production.
The output is what Lasso calls an AI Bill of Materials (AIBOM): a continuously updated record of every agent and AI application in an organization, profiled by model, system prompt, tools, guardrails, policies, and configurations, according to The Hacker News.
Posture Management and Security Graphs
Once the inventory exists, the platform builds a security graph mapping every agent’s connections to LLMs, databases, and APIs. According to the Hacker News report, this visualization makes attack paths visible: security teams can see how a threat actor could move through agent-connected systems before they do.
Static analysis runs against NIST, OWASP, and MITRE frameworks, surfacing misconfigurations with compliance context. Lasso’s framing here is pointed: “The most common AI security failures are not sophisticated attacks. They are misconfigured system prompts, overly permissioned tool access, and guardrails that were never tested against anything adversarial.”
Red Teaming Built for Non-Deterministic Systems
Traditional API testing uses static payload sweeps. Lasso’s red teaming operates in three modes designed for agents that reason, retain memory across turns, and chain tool calls unpredictably. Static attacks draw from a library of more than 300,000 continuously updated payloads with full OWASP LLM and Agentic Top 10 coverage. Dynamic attacks use multi-turn probing to find where an agent’s intent becomes fragile under sustained adversarial pressure. High-agency attacks deploy autonomous adversaries with specialized roles that adapt across turns, exploit context windows and agent memory, and chain tool calls, per The Hacker News.
Why Agent Security Requires Its Own Discipline
Lasso’s launch reflects a broader shift. Agent security is diverging from traditional application security because the fundamental problem is different. As The Hacker News frames it: “Traditional proxies and AI firewalls were built to inspect content: what a user sent, what a model returned. Intent security asks a different question: is this agent doing what it was built to do, in this context, for this user, right now?”
The platform requires no source code access or agent instrumentation, enabling deployment in strict data residency environments. That design choice suggests Lasso is targeting the enterprise procurement bottleneck that Silicon Valley deal lawyer Louis Lehot flagged at the Foley & Lardner CFO Executive Forum this week: major organizations refusing to deploy agents because governance, audit, and liability tooling does not exist.
Lasso is one of at least three agent security developments today. NVIDIA launched the Open Secure AI Alliance with 50+ members building open-source identity and isolation frameworks. And researchers at Accomplish AI disclosed that Anthropic’s Claude Co-Work agent can escape its local sandbox. The convergence suggests agent security operations is maturing into a distinct discipline, with dedicated tooling at every layer of the stack.