The security of AI agents is quietly becoming its own market, and Anaconda has just paid to get into it. On August 4 the company, best known for packaging the Python tools that data scientists live in, announced it had acquired the AI-security startup Enkrypt AI for an undisclosed sum, according to Anaconda's own announcement and coverage at AIwire.
Enkrypt's work sits at two points in an AI system's life. Before a model ships, it runs red-team probes across more than 300 categories of attack, the idea being that a jailbreak or a data leak should surface in testing rather than in front of a customer. Once the model is live, its guardrails sit in the request path and try to block those same failures in real time. Enkrypt also automates the paperwork side, turning frameworks like the NIST AI Risk Management Framework and the EU AI Act into checks a system either passes or does not.
The number that made the case
One figure did a lot of the arguing. Enkrypt says it found 143,000 vulnerabilities across 73 percent of the MCP servers it scanned. MCP, the protocol that lets agents reach out to external tools and data, has spread fast this year, and each connection is a door. A scan finding holes in nearly three quarters of the servers it looked at is the kind of statistic that gets a security acquisition signed.
For Anaconda this is the second such purchase in quick succession, following its acquisition of Kilo Code in July. The through-line is a company that used to sell developer tooling now positioning itself around the harder problem of running that tooling safely once agents are wired into real systems.
It fits a wider mood. Every few weeks brings another reminder that agentic systems can be turned against their owners, from Microsoft's first cyber-defence model to research showing models breaking out of their sandboxes. The tools that make agents useful are the same ones that make them worth attacking, and companies are starting to buy accordingly.
Commentarii · 0