ANALYSIS Three moves in a single week confirm that securing autonomous AI agents has crossed from a feature checkbox into a distinct product and M&A category, with different players staking claims on different layers of the stack.
Why it matters
AI agents are connecting to enterprise applications, browsing the web, and executing code on behalf of organizations. Each of those capabilities creates an attack surface that traditional perimeter security was never designed to cover. The simultaneous appearance of an acquisition, an open-source release, and a new commercial product, all targeting agent security, marks the moment the industry began treating the problem as its own market rather than an afterthought.
The big picture
Cohere signed a letter of intent to acquire AISLE, a San Francisco-based cybersecurity startup that uses AI to detect and fix software vulnerabilities7,8. The deal adds a cybersecurity line to the entity Cohere is assembling through its reported $20 billion merger with Germany's Aleph Alpha. AISLE was founded in 2025 by Ondrej Vlcek, former CEO of Czech antivirus maker Avast, with co-founders from Google DeepMind and Anthropic6. In its first year, AISLE said its AI found more than 500 previously unknown bugs in widely used software, including 12 vulnerabilities in OpenSSL, the free program that encrypts internet traffic for banks, governments, and messaging apps5. Those vulnerabilities were fixed after discovery.
Separately, Perplexity AI open-sourced SPACE, a sandbox it built to isolate AI agents in secure, walled-off virtual environments1. Partners including CoreWeave, Akamai Technologies, and Crusoe are already deploying SPACE and have agreed to offer it to their customers2. Perplexity described the release as a response to "several high-profile incidents that saw autonomous systems escape from secure environments and go on to wreak havoc on third-party websites".
Meanwhile, Radware introduced Model Context Protocol Server Protection, a purpose-built solution for the servers that connect AI agents to enterprise applications, data, and workflows3,4. According to Stacklok's State of Model Context Protocol in Software 2026 survey, cited by Radware, nearly half of software companies surveyed reported using MCP in production, with 19% reporting broad production deployment. Radware's new product joins its existing Agentic AI Protection and LLM Firewall within its cloud security platform.
ANALYSIS Each of the three moves targets a different layer. Perplexity's SPACE addresses the runtime isolation problem: keeping an agent from breaking out of its environment and damaging the host system or the wider internet. Radware's MCP Server Protection sits at the integration layer, guarding the protocol through which agents discover and invoke enterprise tools. Cohere's AISLE acquisition operates at the vulnerability layer, using AI to find and patch the software flaws that agents (or attackers exploiting agents) might reach.
The layering matters because it suggests no single vendor can credibly claim to "secure agents" end to end. Cohere's framing is instructive: "Increasingly capable AI systems are transforming both cybersecurity and the threats organizations face, and defensive capabilities must advance just as quickly," the company said. That language positions security not as a one-time integration but as an ongoing arms race, one that justifies folding a dedicated security team into the core AI platform.
Perplexity's decision to open-source SPACE rather than sell it follows a different logic. By giving the sandbox away and enlisting infrastructure providers such as CoreWeave, Akamai, and Crusoe as distribution partners, Perplexity treats agent isolation as a public good whose ubiquity benefits its own search-agent products. ANALYSIS The contrast with Cohere's acquisition-driven approach and Radware's commercial product launch illustrates how varied the business models around agent security already are.
Radware's emphasis on MCP-specific attacks reflects a newer concern. As MCP adoption grows, the servers that connect agents to enterprise systems "may create a new attack surface that traditional web and DDoS security controls may not have been designed to understand or protect," the company said. ◆ That framing positions MCP servers as the next perimeter, analogous to the API gateway layer that spawned its own security category a decade ago.
What's next
Cohere's AISLE deal remains at the letter-of-intent stage; terms have not been disclosed. Perplexity's SPACE is available to developers starting now. Radware's MCP Server Protection is available as a cloud-based or cloud-augmented service. ◆ With three distinct product categories emerging in a single week, enterprise buyers face an immediate question: which layer of agent security to prioritize first, and whether any vendor can stitch the layers together before the attack surface widens further.