Anaconda pairs agent swarms with autonomous red-team security
Builders using the Anaconda Platform can now run agent swarms alongside autonomous red-team agents, with vetted packages, open-weight models, and governed environments in one place. The combination is meant to let teams ship agentic code quickly while catching security weaknesses before code reaches production, and it draws on new research from Enkrypt AI showing that 73% of agent tools scanned across MCP servers contained vulnerabilities.
Agent swarms are moving from a niche pattern to standard practice. A recent Anaconda survey of AI-native builders found 63% of respondents moving toward swarm-based development in some form, suggesting that coordinating multiple agents in parallel, sharing context, and managing token costs is becoming a default expectation rather than an experiment.
What Anaconda is launching
The new capabilities are powered by three acquisitions: Kilo Code, Enkrypt AI, and Outerbounds. Together they feed into an AI Dev Factory that covers AI Workspaces, Agent Swarms, AI Security and Guardrails, and AI Orchestration.
- AI Workspaces. AI Workspaces brings agentic tools and trusted components together for builders and data scientists, with model freedom and cost optimization as core constraints.
- Agent Swarms in VS Code. Agent Swarms reach VS Code for the first time through Kilo, the primary AI workspace, coordinating multiple agents in parallel, sharing context, and cutting token costs across the group.
- Local AI development with Kilo Desktop. Local AI development with Kilo Desktop combines software engineering, data science, and secure Python environment management, and gives access to more than 500 AI models, 19,000+ vetted packages, local model execution, and live notebook sessions where humans and agents edit and run cells together. Sign in with ChatGPT also gives users access to every Kilo surface without extra logins or added token costs.
- Source-built packages. Source-built packages add 13,000+ newly vetted AI, ML, and Python packages, extending what Anaconda calls the world’s largest source-built library.
- Model Catalog. Model Catalog provides access to 77 vetted open source models in a curated library, expanding builder choice without sacrificing governance.
- MCP. MCP extends the same trust and governance to agent tool calls through the new Anaconda MCP.
How the new security layer works
The security side rests on a hard number from recent agent research. Enkrypt AI scanned more than 268,210 agent tools across 25,264 MCP servers over four months and found vulnerabilities in 73% of them, a finding that frames every other security claim in this release.
- Red-teaming. Red-teaming challenges models, agents, and MCPs across 300+ attack categories, adapting in real time to surface weaknesses during deployment and production.
- Guardrails. Guardrails expand runtime protection, approving, modifying, or blocking risky behavior across agents, tools, RAG, and MCP.
- Agent Incident Registry. Agent Incident Registry gives enterprises a verified, source-backed record of publicly reported agent incidents, described by Anaconda as an industry first that lets buyers confirm security claims beyond vendor statements.
How orchestration closes the development-to-production loop
The orchestration layer moves AI from development through testing and into production.
- Orchestration workflows. Orchestration workflows bring native access to AI Artifacts, so governed packages, models, and security and license policies move with each workflow from development through production.
- Interactive inference. Interactive inference lets teams deploy models straight from the Anaconda model catalog and query them for results, with no separate deployment step.
- Reproducible container images. Reproducible container images are built automatically by FastBakery, which compiles conda and PyPI dependencies, including native libraries, into images pip-only tools cannot produce.
Why enterprises are treating agent autonomy as a top risk
Omdia research found that 72% of organizations rank managing the growing autonomy of AI agents as critical or very important to their AI strategy, which is why the new platform pairs every builder capability with a corresponding governance control. The release is built on a simple premise: as AI agents take on more builder tasks and gain access to enterprise tools and data, security testing and runtime controls have to move at the same pace, and visibility into which agent acted, what data it accessed, which tool it called, and whether it stayed within its assigned authority is now a baseline requirement.
What builders are seeing in early use
One early user of Kilo has been running solo prototyping, spec-writing, and building inside a single workflow, and is now testing agent swarms to coordinate several of those workstreams at once. The result, by their account, is turning what one developer does in a day into what a small team produces, with the practical limit becoming how much one person can handle rather than how much code one person can write.
A short note for buyers comparing platforms: when an article focuses on how visible a business is across agents and AI search engines, an AI-Audit and AI Scan can show what an AI agent actually finds on a website, including the listings and citations behind it.
FAQ
What did Anaconda announce on October 6, 2026?
Anaconda announced new capabilities across the Anaconda Platform that pair agentic development with autonomous security testing, including agent swarms, autonomous red-team agents, and governed packages and models. The release is powered by recent acquisitions of Kilo Code, Enkrypt AI, and Outerbounds.
How vulnerable are current agent tools, according to Enkrypt AI?
Enkrypt AI scanned more than 268,210 agent tools across 25,264 MCP servers over four months and found vulnerabilities in 73% of them, a finding Anaconda is using to frame its own security layer.
What does the Agent Incident Registry actually do?
The Agent Incident Registry gives enterprises a verified, source-backed record of publicly reported agent incidents, described by Anaconda as an industry first that offers an independent way to confirm security claims beyond vendor statements.
Related coverage
BizScoreAI
BizScoreAI scores how visible a business is to AI search and shows what its listing looks like to the engines people ask. Open the AI visibility scan.
This article summarizes reporting from helpnetsecurity.com.