The next enterprise data leak won't come from a hacked database. It will come from a developer pasting proprietary code into ChatGPT. Or a finance analyst uploading a spreadsheet to Claude. ADRI Security, a new Y Combinator-backed startup, is betting that the only way to stop it is to intercept the data before it ever leaves the employee's machine [adrisecurity.ai, home].
Its product is an on-device agent that inspects prompts and file uploads headed to AI chats, coding assistants, and MCP servers. The claim is local detection and redaction of sensitive information, a model that promises to work across VPNs, proxies, and even fully offline, air-gapped deployments [adrisecurity.ai, home]. For security teams watching the uncontrolled spread of AI tool usage, it's a pitch for visibility and enforcement where traditional network perimeters have failed.
The wedge of local control
ADRI's core bet is on architecture. By processing prompts locally on the endpoint, the company argues it can apply data-loss-prevention (DLP) rules without routing sensitive text through a cloud service, maintaining performance for the user and control for the enterprise. The stated targets are both sanctioned tools like GitHub Copilot and the 'shadow AI' applications IT never approved [adrisecurity.ai, home]. This positions the product not as a gateway to block AI, but as a guardrail to enable its safer use.
The technical promise extends to specialized environments. The company highlights support for Model Context Protocol (MCP) servers, which are becoming a standard conduit for connecting AI agents to tools and data sources. Securing this layer is a forward-looking move, aiming to embed controls directly into the emerging plumbing of AI-augmented development [adrisecurity.ai, home].
The founder and the early backing
The venture is led by Darshita Chaturvedi, who is listed as the CEO of ADRI and is also the co-founder and CEO of Atri Labs, a company behind an open-source web development framework [Tracxn, 2026] [Crunchbase]. Her background includes several years in machine learning and data science roles at institutions like MIT, BlackRock, and Credit Suisse [GitNation] [rocketreach.co]. This dual track,building developer tools at Atri Labs while launching an AI security startup,suggests a focus on the engineering end-user as the primary buyer.
Early financial support is light but includes notable programmatic investors. The company has participated in Y Combinator, a standard launchpad for ambitious software bets. According to PitchBook data, ADRI has disclosed approximately $35,000 in total funding, with investors including Flat6Labs, New Zealand Growth Capital Partners, and the Aspire NZ Seed Fund [PitchBook, 2025]. The round size is characteristic of a very early pre-seed or accelerator note.
Disclosed Funding | 35 | K USD
Where the wheels could come off
The ambition is clear, but the path is crowded with technical and commercial hurdles. The company is pre-launch, currently gathering names on a waiting list, which means its core detection efficacy and performance impact are unproven at scale [adrisecurity.ai]. The competitive landscape for AI security is also rapidly formalizing, with established cloud security players and a wave of startups layering controls at the network, SaaS application, or model level.
The company will need to prove several key points to enterprise buyers:
- Detection accuracy. False positives that block legitimate work will cripple developer productivity and lead to immediate uninstalls.
- Performance tax. A local agent that slows down coding or chat interactions will struggle for adoption, especially among the engineers it aims to protect.
- Deployment complexity. Selling a new endpoint agent requires convincing security, compliance, and engineering teams, a classic cross-functional sale with a long cycle.
Furthermore, the founder's concurrent leadership of Atri Labs raises a natural question about bandwidth and strategic focus, though it also provides deep insight into the developer workflow the product aims to secure.
The next twelve months
The immediate milestone is moving from waitlist to general availability. Success will be measured by the first public case studies: named engineering or security teams at mid-to-large enterprises that have deployed the agent and can report a measurable reduction in policy violations without a revolt from their developers. Given the Y Combinator affiliation and the small disclosed pre-seed, a larger seed round to fund this go-to-market push is a likely near-term target.
The $35,000 note from Flat6Labs and NZGCP is a start, but the real check needs to come from a venture firm that believes the endpoint is the only defensible ground in the coming AI security wars. The question for Chaturvedi and her team is whether they can demonstrate enough early technical proof and buyer interest to convince that lead investor to write a multimillion-dollar check within the next year.
Sources
- [adrisecurity.ai, retrieved 2026] ADRI Security Home Page | https://adrisecurity.ai/home/
- [Tracxn, 2026] Adri - 2026 Company Profile, Team, Funding & Competitors | https://tracxn.com/d/companies/adri/__Y1SA1s1-_ULs39_HkBVZwLDquqphjJmtCtIMNZ-23-8
- [Crunchbase, retrieved 2026] Darshita Chaturvedi - Co-Founder and CEO @ Atri Labs | https://www.crunchbase.com/person/darshita-chaturvedi-3e48
- [GitNation, retrieved 2026] Darshita Chaturvedi's talks, articles, workshops, certificates | https://gitnation.com/person/darshita_chaturvedi
- [PitchBook, 2025] ADRI 2025 Company Profile: Valuation, Funding & Investors | https://pitchbook.com/profiles/company/265301-38
- [rocketreach.co, retrieved 2026] Darshita Chaturvedi Email & Phone Number | Adri AI Co-Founder and CEO Contact Information | https://rocketreach.co/darshita-chaturvedi-email_52196353