Builders shipping AI applications can now run agent swarms and continuous security testing inside one platform, following a major update to the Anaconda Platform that ties agentic development, vetted packages, and runtime guardrails together. The expansion addresses a problem most teams hit as agents gain more access to enterprise data: security and visibility have not kept pace with what the agents can do.
What the new platform actually adds
Three capability groups form the backbone of the release. AI Workspaces, Agent Swarms, and a broader Model Catalog give builders more tools and models to choose from. AI Security and Guardrails add red‑teaming and runtime controls. AI Orchestration wraps it all in reproducible environments that move from development into production.
Agent Swarms reach VS Code for the first time through Kilo, the primary AI workspace. Multiple agents coordinate in parallel, share context, and optimize token costs. Kilo Desktop brings local software engineering, data science, and Python environment management into one place, with access to more than 500 AI models, more than 19,000 vetted packages, local model execution, and live notebook sessions where people and agents can edit and run cells together. A new Sign in with ChatGPT option gives users access to every Kilo surface without extra logins or added token costs.
The package and model layers grow significantly. Source‑built packages add more than 13,000 newly vetted AI, machine learning, and Python packages, extending what Anaconda calls the largest source‑built library. The Model Catalog now offers 77 open source models in a curated library, giving builders more choice without dropping governance.
MCP governance is a notable addition. Anaconda MCP extends the same trust and governance the platform applies to packages to agent tool calls, covering the requests agents make to outside tools and data.
How autonomous red‑team testing works
AI Security and Guardrails add two layers. Red‑teaming challenges models, agents, and MCPs across more than 300 attack categories, adapting in real time to surface weaknesses in deployment and production. Guardrails approve, modify, or block risky behavior across agents, tools, RAG, and MCP at runtime.
An Agent Incident Registry gives enterprises a verified, source‑backed record of publicly reported agent incidents, framed as an industry‑first offering that lets organizations check security statements independently rather than relying on vendor claims.
Orchestration that travels from dev to production
AI Orchestration focuses on moving AI reliably from development through testing into production. Orchestration workflows bring native access to AI Artifacts, providing governed packages, models, and security and license policies that travel with each workflow. Interactive inference lets teams deploy models directly from the model catalog and query them for results without a separate deployment step. Reproducible container images are built automatically by FastBakery, which compiles conda and PyPI dependencies, including native libraries, into images that pip‑only tools cannot produce.
Why the timing matters
Anaconda’s recent survey of AI‑native builders showed strong momentum behind agent swarms, with 63% of respondents moving toward the technology in some form, a clear signal that swarms are quickly becoming standard practice.
The need for security at the same pace shows in independent scanning of more than 268,210 agent tools across 25,264 MCP servers over four months, which found vulnerabilities in 73% of them. Builders need visibility into which agent acted, what data it accessed, which tool it called, and whether it stayed within its assigned authority.
Research from Omdia found that 72% of organizations rank managing this growing autonomy as critical or very important to their AI strategy, framing security and visibility as the factors separating enterprises that scale AI successfully from those that stall.
What’s behind the acquisitions
The new release draws on Anaconda’s purchases of Kilo Code, Enkrypt AI, and Outerbounds. The three acquisitions feed directly into the Workspaces, Security, and Orchestration layers, respectively, and the release pulls them together as a single AI Dev Factory for enterprises.
For builders, the practical takeaway is that agent swarms, red‑team testing, and governed model deployment now sit on the same surface, so security checks can run continuously rather than at the end of a build cycle. For organizations, the addition of MCP governance and an incident registry addresses two of the biggest gaps flagged by independent research: tool‑call trust and independent verification of security statements.
FAQ
What is Anaconda’s agent swarm update?
Anaconda has combined agent swarms with autonomous red‑team security testing on the Anaconda Platform. Agent Swarms now reach VS Code through Kilo, coordinating multiple agents in parallel, while new red‑team agents run continuous security tests across more than 300 attack categories.
How many MCP tools and servers were scanned in the security research?
Independent scanning covered more than 268,210 agent tools across 25,264 MCP servers over four months and found vulnerabilities in 73% of them.
What do the new AI Security and Guardrails include?
Red‑teaming challenges models, agents, and MCPs across more than 300 attack categories, and Guardrails approve, modify, or block risky behavior across agents, tools, RAG, and MCP at runtime. An Agent Incident Registry provides a verified, source‑backed record of publicly reported agent incidents.
This article summarizes reporting from helpnetsecurity.com.