The list. Bold Security made the 2026–2027 Fortune Cyber 60 in the early-stage group. Fortune’s profile lists it as founded in 2024, based in Tel Aviv, with 50 employees and CEO Nati Hazut. It only came out of stealth on March 12, 2026. Bold says this is its first year on the list. Hazut has done this before: he founded Polyrize, a cloud data-access company that Varonis bought, and Bold’s product chief, Hadar Krasner, was Polyrize’s CPO.
The problem, in plain words. DLP, or data loss prevention, is software meant to stop sensitive data from leaving the company. Most of it still works by pattern matching. It knows what a credit card number looks like. It does not know that a spreadsheet is next quarter’s financial model or that a folder is your source code. So it misses the things that matter and flags the things that don’t, and most teams end up running it in monitor-only mode. Now AI has moved work back onto the laptop: desktop assistants, coding agents, command-line tools and connections over MCP, the protocol AI apps use to reach files and other tools. A browser extension or a network gateway never sees most of that traffic.
The unique part: the AI runs on the device. Bold puts a lightweight agent on each endpoint, and that agent hosts several small language models. One shared model does most of the heavy work once, and smaller expert models produce the labels: regulated data, type of content, type of business document, how confidential it is. Customers can add classifiers for their own data from a handful of examples. Because there is no cloud round trip, Bold can decide before the action finishes: allow, coach the user, ask for a reason, redirect or block. It reads the process tree, the chain of programs behind an action, to tell a person from an AI agent. Content is analyzed on the laptop. Bold says encrypted evidence goes to the customer’s own cloud with the customer’s key, and Bold receives masked metadata only. That answers two objections at once. Speed is the difference between preventing a leak and filing an alert about it. Privacy is the reason legal teams kill DLP projects.
The smart positioning move. Bold does not try to replace the tools that map data at rest. It pulls in Microsoft Information Protection labels, Google Drive labels and labels from DSPM vendors (data security posture management tools that find sensitive data in storage) such as Cyera, Sentra and Varonis. It wants to be the enforcement point on the device, not another inventory.
The proof, and its limits. On its site, Bold quotes Teza Mukkavilli, a CIO and CISO, saying his team rolled Bold out to more than 3,000 endpoints and saw an 85% reduction in AI incidents within two weeks. Bold also claims 95% faster investigations. These are a customer testimonial and company claims, not audits. Ask what counted as an “AI incident” and whether the starting point was a tool that only watched. Also ask what the models cost the laptop in CPU and battery on your oldest hardware, not on the demo machine.
The financial view. Fortune reports $40 million in funding, with Bessemer Venture Partners, Picture Capital, Red Dot Capital Partners, Cerca and Wiz co-founders Assaf Rappaport and Roy Reznik as angels. Ctech reported a $12 million seed in 2024 and a $28 million Series A in March 2026, which Red Dot says it led. Valuation and revenue are Undisclosed. Fortune does not list Lightspeed as an investor. Headcount was 24 in March, per Ctech, and Fortune lists 50 now.
Why it’s winning now. In Lightspeed’s survey of 200 CISOs, 42% name shadow AI, meaning employees using AI tools nobody approved, among their top threats. Data leaking through AI tools was the second most common thing CISOs wished a startup would solve. The big buyers agree. Ctech reports Cyera paid about $50 million for Genie Security, a five-month-old endpoint data-leak startup. Varonis pushed its Atlas line onto the endpoint today. The endpoint is back as the place data control happens. Bold’s bet is that it only works if the decision happens on the device. The danger is the platform squeeze. CrowdStrike and Microsoft already own an agent on most corporate laptops, and “good enough, already installed” wins plenty of deals.
For practitioners. Run it in block mode on one high-risk team, finance or engineering, for a month. Count how often users override the coaching. That number tells you more than any demo.
Would you trust a model on the laptop to block a file in real time, or does your DLP still live in monitor-only mode?