Every AI Security Startup Got Acquired. Now What?

Three independent AI security companies. Three enterprise acquisitions. Over $1.2 billion spent. And your AI systems are still not secure.

In under 18 months, every major independent AI security startup disappeared into an enterprise vendor:

• Robust Intelligence was acquired by Cisco for approximately $400 million (October 2024). It became the foundation of Cisco AI Defense.
• Protect AI went to Palo Alto Networks for an estimated $500-700 million (July 2025). It now powers Prisma AIRS.
• Lakera was picked up by Check Point for roughly $300 million (Q4 2025). It anchors their new AI Security Center of Excellence.

That is approximately $1.2 to $1.4 billion in acquisitions targeting a single problem: securing AI systems in production.

The enterprise security vendors clearly believe AI security matters. They should. The question is whether their product approach actually solves the problem for most companies.

The product trap

Here is what happens when an independent security startup gets absorbed by a $100 billion platform vendor:

The price goes up. What used to be a focused tool becomes a module inside an enterprise security suite. The minimum spend goes from "talk to sales" to "talk to sales and also buy the rest of the platform." Small and mid-sized companies get priced out.

The scope narrows. Enterprise vendors optimize for their existing customer base. Cisco sells to Cisco shops. Palo Alto sells to Palo Alto shops. Check Point sells to Check Point shops. If your stack does not align, tough luck.

The focus shifts. Independent startups obsess over one problem. Acquisitions dilute that focus. The Protect AI team that used to wake up every morning thinking about ML supply chain attacks now has to think about how their features integrate with 47 other Palo Alto products.

None of this is inherently bad. Enterprise consolidation is a natural market cycle. But it creates a gap.

The gap nobody is filling

The AI security market just lost its independent specialists. Meanwhile, the threat environment got worse.

The numbers are stark. According to the 2026 State of AI Agent Security Report from Gravitee, 80% of organizations are now deploying AI agents. But only 47% of those agents are actively monitored or secured. That means more than half of deployed AI agents operate with zero security oversight.

It gets worse:

• 88% of organizations confirmed or suspected AI-related security incidents this year
• Only 14% have achieved full IT and security approval for their entire agent fleet
• 82% of executives believe their existing policies protect them. Only 22% of teams treat agents as independent identities

This is not a theoretical risk. OpenClaw, the AI agent framework with 135,000 GitHub stars, had over 21,000 exposed instances with critical vulnerabilities including CVE-2026-25253 (CVSS 8.8), plus 341 malicious skills in its marketplace. Microsoft 365 Copilot had the EchoLeak vulnerability (CVE-2025-32711) allowing zero-click data exfiltration through prompt injection. In the first 60 days of 2026, the MCP ecosystem saw 30 CVEs related to tool poisoning attacks.

Enterprise products address some of this. But products cannot replace judgment. They cannot walk into your specific architecture, understand your specific threat model, and tell you where the actual risks are versus where the theoretical risks live.

That requires humans. Specifically, humans who do this work every day and are not trying to upsell you a platform license.

The consulting opportunity is massive

The market just validated itself to the tune of $1.2 billion. And then immediately created a service vacuum.

Consider what mid-market companies actually need:

AI system inventories. The Cloud Security Alliance found that only 21% of organizations maintain a real-time registry of their AI agents, with the EU AI Act high-risk deadline approaching in August 2026. No product auto-discovers shadow AI deployments in your organization. That takes interviews, architecture reviews, and documentation work.

Red teaming that counts. Cyber insurance carriers now require AI-specific security riders. They want documented adversarial red-teaming evidence, model-level risk assessments, and alignment with recognized risk frameworks. The AI red teaming services market is projected to hit $2.26 billion in 2026, growing at nearly 29% annually. You cannot get an insurance-qualifying red team report from a SaaS dashboard. You need someone to actually attack your systems and write up what they found.

EU AI Act compliance documentation. Finland activated AI Act enforcement powers in January 2026 and other EU states are following. Romania has not even designated a competent authority yet, which means companies operating there are essentially on their own. The documentation requirements (risk assessments, conformity assessments, post-market monitoring plans) require deep knowledge of both the regulation and your technical architecture.

Vendor-neutral assessments. When your security vendor is also your AI security vendor, who audits the auditor? Companies increasingly need independent eyes on their AI infrastructure, not another tool from the same vendor that sold them the firewall.

What actually works

We work on AI security and MLOps at DeviDevs. Here is what we see actually moving the needle for companies trying to secure their AI systems:

Start with the inventory. You cannot secure what you cannot see. Map every AI system, every model, every agent, every API endpoint that touches ML inference. This is boring work. It is also the single most impactful thing you can do before August 2026.

Red team your agents, not just your models. Model security (adversarial examples, data poisoning, model theft) gets all the attention. But in 2026, the real risk is agent security: what happens when your AI agent has tool access, can execute code, or can interact with external systems. Red team the agent's entire execution boundary, not just the model weights.

Get your documentation insurance-ready. If you have cyber insurance or plan to renew, ask your carrier about AI security riders. Then work backwards from their requirements. You will likely need documented red team results, a risk management framework mapping, and evidence of ongoing monitoring. Build these artifacts now, not during your renewal crunch.

Do not assume your platform vendor has you covered. Cisco AI Defense, Palo Alto Prisma AIRS, and Check Point's AI security stack are all good products. They are also all optimized for enterprise customers with six-figure security budgets. If that is not you, or if you need an independent assessment alongside your vendor tools, find a specialist.

The bottom line

The acquisition wave proves that AI security is a billion-dollar problem. It also proves that the market is consolidating around enterprise product plays that leave most companies underserved.

The independent specialists are gone. The enterprise products are expensive and vendor-locked. The threat environment is accelerating faster than most security teams can track.

If your organization is deploying AI agents (and statistically, you probably are), the question is not whether you need AI security. The market already answered that with $1.2 billion.

The question is who is actually going to do the work.

About DeviDevs: We build ML platforms, secure AI systems, and help companies comply with the EU AI Act. devidevs.com