AI agent security is being absorbed into the big security platforms. If you already pay an EDR, XDR, or SSE vendor, you will soon be offered agent discovery, red teaming, guardrails, and observability as one more module. The practical question for IT leaders is not whether consolidation is happening. It is when the platform is the right buy, when a specialist still wins, and how to keep your options open either way.
On November 3, 2026, at Fal.Con Europe, CrowdStrike announced agreements to acquire two AI security startups: Fiddler AI and Dreadnode. According to CrowdStrike's announcement, Fiddler's technology will be sold as Falcon Observe. It covers pre-deployment evaluation of accuracy, task completion, and policy compliance, plus production tracing of agent decisions, reasoning paths, and tool calls. It also flags drift, hallucinations, and unintended actions, and keeps audit records. CrowdStrike expects support for OpenTelemetry along with LangGraph, Amazon Bedrock, AWS Strands, and Google ADK.
Dreadnode brings automated AI security testing. It runs more than 100 attack strategies, including jailbreaks, prompt injection, multi-turn attacks, and attacks on agent reasoning loops and MCP servers. Findings map to the OWASP Top 10 for LLMs, OWASP agentic security guidance, MITRE ATLAS, and the NIST AI RMF, and feed into CrowdStrike's Falcon Guardian. Both deals are expected to close and reach general availability in CrowdStrike's fiscal fourth quarter of 2027. Terms were not disclosed, though the Wall Street Journal reported about $495 million combined.
CrowdStrike's framing is worth remembering: agents "don't have to be compromised to create risk." An agent with too much access can do damage simply by following its instructions badly.
This is part of a longer pattern. Cisco bought Robust Intelligence in 2024. In 2025, Palo Alto Networks announced its acquisition of Protect AI, SentinelOne acquired Prompt Security, Check Point acquired Lakera, and F5 acquired CalypsoAI. Nearly every major security platform now has, or is assembling, an AI security story.
AI agent security breaks into four capability layers: discovery, pre-deployment testing, runtime guardrails, and observability. Any vendor you evaluate, platform or specialist, should be scored against all four.
Consolidation changes who sells you these layers, not what the layers need to do. The main shifts are practical:
Gartner analyst Apeksha Kaushik, quoted by CSO Online, cautioned enterprises to distinguish announced plans from demonstrated capability and to ask about pricing, packaging, migration, service continuity, data handling, telemetry retention, and API access. That advice applies to every vendor in this market, not just one.
Buy from your platform when it already runs well in your environment and can demonstrate the layers you need today. Buy a specialist when you need depth, breadth across many stacks, or independence. Use this table as a starting point.
| Factor | Lean toward your platform vendor | Lean toward a specialist |
|---|---|---|
| Where agents run | Mostly on endpoints, clouds, and identities the platform already covers | Across many clouds, frameworks, and SaaS tools the platform does not see |
| Team capacity | Small security team that needs one console | Dedicated AI or AppSec engineers who will use advanced features |
| Capability maturity | Features are generally available and demonstrated in your environment | Platform offering is still a roadmap or awaiting acquisition close |
| Testing depth | Standard red team coverage mapped to OWASP and MITRE ATLAS is enough | You build custom agents with complex tool chains that need tailored attacks |
| Contract position | Renewal is near and bundling gives leverage | You want to avoid deepening dependence on one vendor |
| Regulatory needs | Platform meets your data residency and retention rules | You need specific residency, retention, or audit formats the platform lacks |
Many organizations land on a hybrid: the platform for discovery, runtime policy, and SOC correlation, plus a specialist or in-house practice for deep testing of high-risk custom AI agents.
When a vendor is buying or being bought, ask for evidence, not slides. Get written answers to these:
Own your data and your rules, and let vendors operate on them. Lock-in in AI security comes less from the console and more from the inventory, policies, and history that accumulate inside it.
Governance is cross-functional, and security is one owner among several. As the Gartner commentary noted, security teams do not become sole owners of AI agent governance just because the tooling now lives in their platform.
A workable model assigns clear roles. Security owns controls, monitoring, and incident response. Business owners own each agent's purpose and acceptable outcomes. Data owners approve what agents can read and write. Legal and compliance set retention, disclosure, and regulatory requirements. Engineering owns build quality and testing. A small council with a named executive sponsor reviews new high-risk agents and settles disputes. Our AI consulting team helps organizations stand this up without creating a bottleneck.
In 90 days you can go from scattered tools to a deliberate buying decision with portable foundations.
If you want a structured starting point, an AI assessment covers the inventory and gap analysis, and our AI security and governance practice can run the vendor comparison with you. Organizations running agents on dedicated infrastructure can also review how controls apply to hosted AI environments.
Consolidation makes platform-based AI agent security easier to buy, but not automatically better. Score every option against the four layers, insist on demonstrated capability over roadmaps, keep your inventory, policies, telemetry, and tests portable, and share governance across the business. Do that and you can take the convenience of a platform without giving up your leverage.
Often yes, if you already run that platform well, your agents live mostly in environments it covers, and the vendor can show working capability today rather than a roadmap. Buy a specialist when you need depth the platform lacks, run agents across many clouds and frameworks, or want leverage while acquisitions settle.
On November 3, 2026, CrowdStrike announced agreements to acquire Fiddler AI for agent observability and Dreadnode for automated AI red teaming. Both are expected to close and reach general availability in CrowdStrike's fiscal fourth quarter of 2027. Terms were not disclosed, though the Wall Street Journal reported about $495 million combined.
Keep your agent inventory, policies, and test cases in formats you own, require OpenTelemetry export for agent traces, and confirm API access to your data and findings. If you can move your telemetry and rules to another tool in weeks, the platform is a convenience rather than a trap.
No single team. Security owns controls and threat response, but business owners, data owners, legal, compliance, and the engineering teams that build agents all share accountability. A small cross-functional council with a named executive sponsor works better than handing it all to the SOC.
Ask what ships today versus what is planned, the integration timeline, whether current contracts and pricing are honored, where data is stored and for how long, how telemetry can be exported, and what happens to features that overlap with the acquirer's existing products.
Infonaligy is based in Dallas–Fort Worth and helps organizations nationwide evaluate and secure AI agents remotely; see all locations we serve.
We help IT leaders inventory their agents, test them, and compare platform and specialist options against the same criteria, so the decision rests on evidence and stays portable.