Rick Caccia, CEO and Co-Founding father of WitnessAI, has intensive expertise in launching safety and compliance merchandise. He has held management roles in product and advertising at Palo Alto Networks, Google, and Symantec. Caccia beforehand led product advertising at ArcSight by means of its IPO and subsequent operations as a public firm and served as the primary Chief Advertising and marketing Officer at Exabeam. He holds a number of levels from the College of California, Berkeley.
WitnessAI is growing a safety platform centered on making certain the secure and safe use of AI in enterprises. With every main technological shift—comparable to internet, cellular, and cloud computing—new safety challenges emerge, creating alternatives for business leaders to emerge. AI represents the subsequent frontier on this evolution.
The corporate goals to determine itself as a pacesetter in AI safety by combining experience in machine studying, cybersecurity, and large-scale cloud operations. Its group brings deep expertise in AI improvement, reverse engineering, and multi-cloud Kubernetes deployment, addressing the crucial challenges of securing AI-driven applied sciences.
What impressed you to co-found WitnessAI, and what key challenges in AI governance and safety had been you aiming to resolve?
After we first began the corporate, we thought that safety groups can be involved about assaults on their inside AI fashions. As an alternative, the primary 15 CISOs we spoke with mentioned the other, that widespread company LLM rollout was a good distance off, however the pressing downside was defending their staff’ use of different folks’s AI apps. We took a step again and noticed that the issue wasn’t heading off scary cyberattacks, it was safely enabling corporations to make use of AI productively. Whereas governance possibly much less attractive than cyberattacks, it’s what safety and privateness groups really wanted. They wanted visibility of what their staff had been doing with third-party AI, a method to implement acceptable use insurance policies, and a method to defend information with out blocking use of that information. In order that’s what we constructed.
Given your intensive expertise at Google Cloud, Palo Alto Networks, and different cybersecurity corporations, how did these roles affect your method to constructing WitnessAI?
I’ve spoken with many CISOs over time. Some of the widespread issues I hear from CISOs immediately is, “I don’t need to be ‘Physician No’ on the subject of AI; I need to assist our staff use it to be higher.” As somebody who has labored with cybersecurity distributors for a very long time, it is a very completely different assertion. It’s extra paying homage to the dotcom-era, again when the Internet was a brand new and transformative know-how. After we constructed WitnessAI, we particularly began with product capabilities that helped clients undertake AI safely; our message was that these items is like magic and naturally everybody needs to expertise magic. I feel that safety corporations are too fast to play the worry card, and we wished to be completely different.
What units WitnessAI aside from different AI governance and safety platforms out there immediately?
Nicely, for one factor, most different distributors within the house are centered totally on the safety half, and never on the governance half. To me, governance is just like the brakes on a automobile. If you happen to actually need to get someplace shortly, you want efficient brakes along with a robust engine. Nobody goes to drive a Ferrari very quick if it has no brakes. On this case, your organization utilizing AI is the Ferrari, and WitnessAI is the brakes and steering wheel.
In distinction, most of our rivals give attention to theoretical scary assaults on a company’s AI mannequin. That may be a actual downside, nevertheless it’s a unique downside than getting visibility and management over how my staff are utilizing any of the 5,000+ AI apps already on the web. It’s quite a bit simpler for us so as to add an AI firewall (and we’ve got) than it’s for the AI firewall distributors so as to add efficient governance and threat administration.
How does WitnessAI stability the necessity for AI innovation with enterprise safety and compliance?
As I wrote earlier, we consider that AI needs to be like magic – it could possibly assist you to do wonderful issues. With that in thoughts, we predict AI innovation and safety are linked. In case your staff can use AI safely, they’ll use it usually and you’ll pull forward. If you happen to apply the everyday safety mindset and lock it down, your competitor gained’t do this, and they’ll pull forward. Every part we do is about enabling secure adoption of AI. As one buyer instructed me, “These things is magic, however most distributors deal with it prefer it was black magic, scary and one thing to worry.” At WitnessAI, we’re serving to to allow the magic.
Are you able to discuss concerning the firm’s core philosophy concerning AI governance—do you see AI safety as an enabler slightly than a restriction?
We recurrently have CISOs come as much as us at occasions the place we’ve got offered, and so they inform us, “Your rivals are all about how scary AI is, and you’re the solely vendor that’s telling us learn how to really use it successfully.” Sundar Pichai at Google has mentioned that “AI may very well be extra profound than fireplace,” and that’s an attention-grabbing metaphor. Hearth may be extremely damaging, as we’ve got seen not too long ago. However managed fireplace could make metal, which accelerates innovation. Generally at WitnessAI we speak about creating the innovation that permits our clients to soundly direct AI “fireplace” to create the equal of metal. Alternatively, in the event you assume AI is akin to magic, then maybe our purpose is to offer you a magic wand, to direct and management it.
In both case, we completely consider that safely enabling AI is the purpose. Simply to offer you an instance, there are lots of information loss prevention (DLP) instruments, it’s a know-how that’s been round perpetually. And folks attempt to apply DLP to AI use, and possibly the DLP browser plug in sees that you’ve typed in a protracted immediate asking for assist along with your work, and that immediate advertently has a buyer ID quantity in it. What occurs? The DLP product blocks the immediate from going out, and also you by no means get a solution. That’s restriction. As an alternative, with WItnessAI, we are able to determine the identical quantity, and silently and surgically redact it on the fly, after which unredact it within the AI response, so that you just get a helpful reply whereas additionally protecting your information secure. That’s enablement.
What are the largest dangers enterprises face when deploying generative AI, and the way does WitnessAI mitigate them?
The primary is visibility. Many individuals are stunned to study that the AI software universe isn’t simply ChatGPT and now DeepSeek; there are actually 1000’s of AI apps on the web and enterprises take in dangers from staff utilizing these apps, so step one is getting visibility: which AI apps are my staff utilizing, what are they doing with these apps, and is it dangerous?
The second is management. Your authorized group has constructed a complete acceptable use coverage for AI, one which ensures the security of buyer information, citizen information, mental property, in addition to worker security. How will you implement this coverage? Is it in your endpoint safety product? In your firewall? In your VPN? In your cloud? What if they’re all from completely different distributors? So, you want a method to outline and implement acceptable use coverage that’s constant throughout AI fashions, apps, clouds, and safety merchandise.
The third is safety of your individual apps. In 2025, we are going to see a lot quicker adoption of LLMs inside enterprises, after which quicker rollout of chat apps powered by these LLMs. So, enterprises want to ensure not solely that the apps are protected, but in addition that the apps don’t say “dumb” issues, like suggest a competitor.
We tackle all three. We offer visibility into which apps individuals are accessing, how they’re utilizing these apps, coverage that’s based mostly on who you might be and what you are attempting to do, and really efficient instruments for stopping assaults comparable to jailbreaks or undesirable behaviors out of your bots.
How does WitnessAI’s AI observability characteristic assist corporations monitor worker AI utilization and stop “shadow AI” dangers?
WitnessAI connects to your community simply and silently builds a catalog of each AI app (and there are actually 1000’s of them on the web) that your staff’ entry. We inform you the place these apps are situated, the place they host their information, and so on so that you just perceive how dangerous these apps are. You possibly can activate dialog visibility, the place we use deep packet inspection to look at prompts and responses. We are able to classify prompts by threat and by intent. Intent is likely to be “write code” or “write a company contract.” It’s vital as a result of we then allow you to write intent-based coverage controls.
What position does AI coverage enforcement play in making certain company AI compliance, and the way does WitnessAI streamline this course of?
Compliance means making certain that your organization is following rules or insurance policies, and there are two elements to making sure compliance. The primary is that you will need to be capable of determine problematic exercise. For instance, I must know that an worker is utilizing buyer information in a method which may run afoul of a knowledge safety legislation. We do this with our observability platform. The second half is describing and imposing coverage towards that exercise. You don’t need to merely know that buyer information is leaking, you need to cease it from leaking. So, we we have constructed a singular AI-specific coverage engine, Witness/CONTROL, that permits you to simply construct identification and intention-based insurance policies to guard information, stop dangerous or unlawful responses, and so on. For instance, you’ll be able to construct a coverage that claims one thing like, “Solely our authorized division can use ChatGPT to put in writing company contracts, and in the event that they achieve this, mechanically redact any PII.” Simple to say, and with WitnessAI, simple to implement.
How does WitnessAI tackle considerations round LLM jailbreaks and immediate injection assaults?
We now have a hardcore AI analysis group—actually sharp. Early on, they constructed a system to create artificial assault information, along with pulling in broadly obtainable coaching information units. Consequently, we’ve benchmarked our immediate injection towards the whole lot on the market, we’re over 99% efficient and recurrently catch assaults that the fashions themselves miss.
In follow, most corporations we communicate with need to begin with worker app governance, after which a bit later they roll out an AI buyer app based mostly on their inside information. So, they use Witness to guard their folks, then they activate the immediate injection firewall. One system, one constant method to construct insurance policies, simple to scale.
What are your long-term objectives for WitnessAI, and the place do you see AI governance evolving within the subsequent 5 years?
Up to now, we’ve solely talked a couple of person-to-chat app mannequin right here. Our subsequent part will likely be to deal with app to app, i.e agentic AI. We’ve designed the APIs in our platform to work equally nicely with each brokers and people. Past that, we consider we’ve constructed a brand new method to get network-level visibility and coverage management within the AI age, and we’ll be rising the corporate with that in thoughts.
Thanks for the good interview, readers who want to study extra ought to go to WitnessAI.
