Networks Generation

Veraification Of The Need To Track And Control Those Pesky AI Agents (no spelling mistakes)

Steve Broadhead
Steve Broadhead
Broadband Testing

Increasingly, AI is turning into a Science Fiction movie – the real life version.

Meantime, the fightback has started – from within. For starters, why have one company in this AI era, when you can have two?

Such appears to be the thinking behind my client Cloudbrink launching Veraify “powered by Cloudbrink” – designed as a means of allowing enterprises to adopt AI at scale, securely – to which we all say hallelujah (even Leonard Cohen temporarily rose from his grave in appreciation). Well, the “powered by” logo was good enough for Intel in its day, so why not? And, no, the new “division” of Cloudbrink wasn’t created by AI, but humans of the technology and marketing persuasion methinks.

A key driver behind the platform – which combines endpoint intelligence, AI-aware policy enforcement, data protection and secure high-performance connectivity (that being Cloudbrink’s “thing” as proven by yours truly – report still available here: https://cloudbrink.com/broadband-performance-tests/) was seemingly a combination of AI agents and embedded AI running amuck, unseen by “old school” traffic inspection platforms (and unsecured by lazy AI app developers) and the new, updated and even more dangerous form of shadow IT – welcome to shadow AI (no Hank Marvins involved – keeping the increasingly tenuous musical subplot going).

I was deep in sh… IT the first time it went into shadow mode; disgruntled department heads, fed up of the lead times given by IT bringing their own applications on board and – worse still – modem connections (hello back doors). And the rest. Fortunately, cybercrime was very much its in infancy, otherwise few companies would have survived those times. Now, the threat is a LOT greater, by a factor of about a trillion to one (conservative estimate). Think unsecured agentic AI in supply chains, for example. Scary. Employees are using whatever their favourite flavours of LLM are –  ChatGPT, Claude, Gemini, Copilot, whatever – custom AI apps, launching autonomous agents into the world of global connectivity and… do they tell SecOps? Er… I don’t think so (I don’t think they even know SecOps exists).

So, for these reasons and more, Veraify has entered the building (vague Elvis link there). What, then, does it do? In the proverbial nutshell, Veraify is designed to:

  • Discover Every AI Interaction. Identify sanctioned and unsanctioned AI tools, agents, copilots, and services. Agent-based policy enforcement captures all AI traffic and enforces policy at the source/endpoint. AI traffic cannot bypass the agent, ensuring full visibility, governance controls, and evidence recording for compliance.
  • Protect Sensitive Data Before It Leaves. Prevent employees and AI agents from exposing private information, code, customer data, or intellectual property. Advanced DLP capabilities include a built-in PII catalogue to begin AI protection within minutes and eliminate constant DLP policy updates when PII data types and formats change. Multi-level content inspection monitors and secures PII across AI prompts, documents, or images, including images inside files uploaded to AI.
  • Anticipate and Prepare for New AI. Automated learning and discovery of new AI platforms and agents that are not in the integrated top 100 AI catalogue eliminates the need for customers to continually manually configure the detection logic for new platforms
  • Govern Humans and AI Agents Together. Apply identity, device, data, and AI-aware policies from a unified control plane regardless of the origin.
  • Secure AI Without Sacrificing Performance. Enable AI adoption without forcing users into slow legacy SASE and DLP architectures they then bypass, using the aforementioned Cloudbrink connectivity architecture.

You can read more about the new company here: https://veraify.ai/ but they also carried out an early adopter’s survey, which will be available on their website. Meantime, here is a selection of the key findings from it:

AI has become a business priority

Nearly every organisation surveyed has already started deploying AI in some form.

  • 37.5% are experimenting with public AI tools such as ChatGPT, Claude and Gemini.
  • 31.3% are already building or deploying internal AI applications.
  • 18.8% already have enterprise AI projects in production.
  • Only 6.3% have not formally started using AI.

Security concerns are centred around visibility and governance

The highest concerns identified were:

  • Employees uploading sensitive information into AI services.
  • Shadow AI and unauthorised AI applications.
  • AI agents accessing company data without appropriate controls.
  • Lack of visibility into AI usage.
  • Regulatory and compliance exposure.

Traditional networking and infrastructure issues ranked far lower than AI governance concerns.

AI security is now an executive issue

Over 80% of respondents described AI security as either highly important or an executive priority, showing that AI governance is rapidly moving beyond IT into board-level discussions.

Existing security tools are not considered sufficient

When asked whether existing security platforms could adequately control AI risk:

  • 37.5% were uncertain.
  • 31.3% were not confident.
  • Only 31.3% believed their current security tools were sufficient (Note: I’d be amazed if they were – yours truly).

This highlights a significant gap between AI adoption and AI governance capabilities.

Summing up, then, we are seeing a response from IT to the threat mass AI adoption has created, as noted recently in this space regarding my test work with Atsign, for example. AI may be driven by IT, but it is also capable of destroying it – literally. So, it’s good to see the responsible side of IT fighting back (even if, in some cases it involves using AI – sensibly) or only exists in the first place because of the “AI rush” – there’s artificial intelligence in them, there hills of networks… (or something like that).

In truth, I always hankered back to the “Wild West” days of IT, before it all became too predictable. Now, it seems, thanks to AI, the Wild West is back. Ye-ah!