Product1 publisher3 min readPublished
Crosspoint leads a $72M round into Mind to bring DLP to files AI agents move unattended
Mind Security's Series B funds a product with two controls: block the risky action, or prompt the person who triggered it with the policy they were about to break. Agent traffic under a service account leaves only the block.
The Product Desk · Product desk
What happened
- Mind Security said it raised $72 million to push further into large enterprise accounts with new hires and channel partnerships, in a Series B led by Crosspoint Capital Partners with returning backers YL Ventures and Paladin Capital Group.
- The software locates sensitive files across SaaS applications, endpoints and email, then classifies each one on its content as well as its context.
- Children's Hospital Los Angeles uses it to keep protected health information out of AI tools running on staff devices, according to Chief Information Officer Conrad Band.
- In May the company became the first data security firm accepted into Anthropic's Cyber Verification Program, which lifts default limits on dual-use security work for vetted firms.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint Half of the control model needs a human at the keyboard. Where an agent acts under a service account there is nobody to prompt, so a security team is left with allow or block, and a block interrupts work that was already approved.
- decision Anyone pricing an AI-DLP line item now has to count the actions they want covered by identity type: those carrying a named human, and those running headless.
- cost The implied average contract sits in the hundreds of thousands of dollars a year. A contract that size sets the size of buyer this product is currently built to serve.
- exposure One named deployment puts the control between clinician devices and AI tools holding protected health information, so a missed classification becomes a health-record disclosure.
The prompt is the half that presumes a person. Mind's case for the round is that older DLP products were written for predictable, human-paced workflows. The second half of the case is that the assumption collapses once AI agents begin reading, transforming and sharing company data with nobody in the loop [8]. Even taking both at face value, the coverage for agent traffic is thinner than the pitch. There is nobody to receive the prompt, so what is left is allow or block [7].
Blocking is the expensive control. It stops a workflow mid-run.
The evidence offered for the category claim is mostly Mind's own. The announcement carries no incident data showing older tools missing agent-initiated movement. Research the company published this year found that 65% of enterprises lack confidence in the controls they have over the data feeding their AI systems [9]. "Security leaders are being asked to protect data that moves at AI speed with complex, manual and incomplete tools designed for a slower world," said co-founder and Chief Executive Eran Barak [10]. Buyer-side confirmation in the announcement runs to a single line. Mark DeCarlo, senior manager of security engineering and technology at the National Geographic Society, said the software gives his team a view of "where and when our data interacts with AI agents" [20].
Mind said revenue rose more than 17-fold over the past year and its customer count eightfold, without disclosing the underlying figures [12]. Divide the first multiple by the second and revenue per customer roughly doubled [15]. The doubling points at expansion inside accounts the company already had. Barak described a business with eight-figure revenue and dozens of customers [13]. At the low end of both, $10 million spread across 36 customers, the average is about $278,000 a customer a year [16]. The software now runs across hundreds of thousands of endpoints [14].
The company puts total funding at $112 million [4], counting an $11 million seed in October 2024 and a $30 million Series A in June 2025 [3]. Those two plus this round add to $113 million [5]. The Series B arrives about 15 months after the Series A [24].
Mind's answer to agent risk is also built out of agents. The AI DLP Agents it released in July build custom classifiers, draft policies and investigate incidents, and analysts direct them in plain language through a Model Context Protocol interface [17]. That is the same class of plumbing that puts a model in front of company data in the first place. Mind holds ISO/IEC 42001 certification for AI management systems [22]. Zach Sivertson, a managing director at Crosspoint, cited Mind's progress since the Series A and its run as a top 10 finalist in the 2025 RSAC Innovation Sandbox contest. Those had "further strengthened our conviction in the company", he said [23].
For anyone sizing this against an existing DLP program, the sorting question is what share of the actions you want covered carry a named human identity and what share run headless. For the human share, the prompt is the cheap control, because the user corrects the move and the job continues. For the headless share, the block is the only control left.
What to watch
- Whether Mind or a named customer publishes incident data showing older DLP tools missing agent-initiated data movement.
- Whether the AI DLP Agents that draft policies and investigate incidents get controls applied to them, given they act through a Model Context Protocol interface.
- Whether pricing appears below the roughly $278,000 per customer implied by eight-figure revenue across dozens of accounts. That would signal a move down-market.