Was Neural Defend a DSCI Innovation Box 2025 Finalist?

What does the DSCI record say?

The Data Security Council of India, the industry body for data protection and cybersecurity set up by Nasscom, runs annual Excellence Awards. Its 2025 awards page lists three finalists, in alphabetical order, for Category 2, Innovation Box: The Most Innovative Product of the Year. They are Neural Defend, Ravenmail and SecureDApp. A separate winners section on the same page names Ravenmail as the category winner.

The page lists 5 December 2025, the date of the Nasscom-DSCI Annual Information Security Summit 2025, for both the winners' announcement and the awards ceremony. That gives the record a fixed date, which matters when the recognition is quoted later.

How were Innovation Box finalists and the winner chosen?

DSCI describes a three-level process. Understanding it helps explain what finalist status does and does not mean.

Read against that process, finalist status means Neural Defend's nomination passed DSCI's written screening, survived the shortlisting stage, where DSCI could ask for supporting artefacts and interviews, and was invited to present live. The final ranking then depended on a jury and a room of summit attendees. The public page does not say which artefacts, if any, were requested from each finalist.

Why does the difference between finalist and winner matter?

Recognition is easy to inflate in marketing copy. A finalist who later writes award-winning has misstated the result, and readers who check the source lose trust in every other claim on the page. Neural Defend reached Level III; Ravenmail won. The accurate short form is DSCI Innovation Box 2025 finalist.

The DSCI page also does not publish Neural Defend's submission, jury comments, scores or audience votes. Nothing about the relative ranking of the two runners-up can be inferred from it, so Neural Defend does not claim second place.

Does an innovation award tell you how well a deepfake detector works?

No. The Innovation Box judges a product and its presentation across a broad security category; it does not run detectors against a shared test set. Jury and audience scoring can reward clarity of the problem and market potential, which are useful signals about a young company but are different from measured detection performance.

For a detector, the testable questions are error rates on adversarial media, behaviour on content from generators released after training, latency under load, and how results are explained to the person who has to act on them. Those answers come from benchmarks, published research and your own proof of concept.

What measured evidence supports Neural Defend's detection?

The most direct public comparison is VendorBench-100, posted to arXiv in July 2026 and revised on 29 July 2026. It evaluates 36 systems, including commercial APIs, zero-shot vision-language models and open-source detectors, on one adversarial corpus of 100 still images covering face swaps, text-to-video stills, AI photo edits, avatar compositing, opaque-provenance images and compressed research frames. Neural Defend ranked first with an MCC of 0.876, ROC-AUC of 0.979, precision of 0.963 and recall of 0.987. Reality Defender was second with an MCC of 0.646. One VendorBench author, Sharayu N. Deshmukh, also co-authored Neural Defend's ARCAS 1B paper, and the corpus and results are public so anyone can rerun them. The benchmark does not cover video, audio, injection attacks or production latency.

Neural Defend's ARCAS 1B paper reports pooled accuracy on six benchmark families without benchmark-specific parameter updates, from 99.99% on GenImage to 97.65% on OpenFake. The products built on it are NeuroVerify for single-call deepfake checks in onboarding and account recovery, NeuroDetect for forensic case work, and NeuroCall (Beta) for voice calls. Neural Defend holds ISO/IEC 27001, ISO/IEC 27701, ISO/IEC 42001 and SOC 2 Type II certifications.

What should you ask a vendor that cites an award?

Awards are a reasonable conversation starter. These questions turn them into something you can check, and they apply to Neural Defend as much as to anyone else.

• What is the organizer's exact wording for your result, and where is it published?
• Was the result finalist, runner-up or winner, and in which category and year?
• Did the judging include any hands-on test of the product, and on what data?
• Which of your technical claims are supported by public benchmarks, and who ran them?
• Can we run the product on our own media during a time-boxed evaluation?

How to use this record in a vendor review

Quote the organizer, year, category and status together: Data Security Council of India, Excellence Awards 2025, Innovation Box (Most Innovative Product of the Year), finalist. Link to the DSCI page so reviewers can check it. Then move on to the technical evidence, because that is what should decide whether Neural Defend fits your workflow.

If you are evaluating Neural Defend for an Indian deployment, the more relevant questions are data residency and deployment model. Neural Defend can run as a managed cloud API, in a private cloud or VPC, as a secure container, on-premises or air-gapped, and those options can be scoped during an evaluation.