Recorded as part of the Black Hat India podcast series ahead of the inaugural Black Hat India conference in Bengaluru (October 27–30, 2026), this ~32-minute discussion is moderated by Swati Aruna (Principal Consultant, ICT domain, Frost & Sullivan India) with two members of the Briefings Review Board: Neelu Tripathy (Senior Security Architect, Adobe) and Anant Shrivastava (Founder, Cyfinoid Research). Framed by a reported ₹22,000 crore lost to cyber fraud in 2025 — digital arrest, investment scams, and AI-enabled fraud — the conversation covers where AI genuinely helps defenders, how it is accelerating attackers, why inventory remains the industry’s oldest unsolved problem, and what the review board wants to see in the first-ever Black Hat India Call for Briefings.
Summary
Swati opens by asking both board members where defenders will benefit from AI over the next three years. Neelu Tripathy argues that although AI is “in its infancy,” its core strength is pattern recognition — joining the dots across the enormous volume of logs security teams already collect, automating detection engineering and attack-path work, and enabling real-time threat-intel correlation.
Anant Shrivastava builds on this with two frames. First, acceleration: work that previously required a person of “higher than average intelligence” can now be done by average-intelligence people, because LLMs capture and distribute those patterns. Second, a concept he calls “rented cognition” — most of us are renting intelligence from a service provider — and his prescription is to keep that rented portion minimal, using LLMs to augment existing capability while moving intelligence from non-deterministic models into deterministic systems where defenders get true, repeatable power. Investigation, pattern matching, and triage are the areas he sees LLMs helping most, given the chronic shortage of security manpower.
Asked what has changed most for researchers in two years, Anant describes a development-culture shift: we have moved from being developers to being managers of developers, with software now routinely accepted if output merely “looks good enough” against a set of inputs rather than being verified as correct. On the attacker side, the reaction was simpler — “we got a new toy.” He points to the Shai-Hulud npm worm and a recent release by Myasnikov as examples of malware being open-sourced so others can trivially fork and improve it: older Shai-Hulud exfiltrated stolen content in plaintext (easy for defenders to track), while newer forks encrypt it — so defenders can no longer even see who is being attacked.
Neelu then lays out a three-layer prioritization for organizations over the next “very experimental” couple of years, noting that security follows the business and that a flood of fast-generated, not-necessarily-clean AI code is already arriving:
- Be where the developer is — put guardrails in the IDE (e.g., if they use Cursor, add controls there; if they use Claude, add skills there), and embed checks and scans in CI pipelines — creatively, not just deterministically.
- Arm the enterprise boundary — many security vendors now ship AI-enabled modules that do pattern recognition and back-end correlation on top of detection; use them as a second layer.
- Bake fundamentals into platforms — combine the deterministic controls teams already have with probabilistic AI-driven controls in the platform itself, so developers keep their freedom while baseline security, configuration security, and agentic/AI-development controls come for free, with visibility included.
She closes that segment with the line that anchors the whole episode: “If you cannot see your attack surface, you cannot secure it” — and no AI can fix inventory without visibility. Real-time (if not yet Minority-Report-style proactive) detection is the realistic goal.
Asked about the biggest capability gap, Anant picks up Neelu’s one word — inventory. IT and IT security never had a proper inventory, and even when they have one, they don’t know what to do with it. He cites SBOM as the cautionary tale: the pushback comes mostly from developers — the very people who would benefit most — because the narrative reduced SBOMs to “finding bugs and vulnerable libraries,” creating negative sentiment. His broader prescription is attack-surface reduction: imagine 100 Node.js projects, each pulling ~10,000 dependencies, where 20,000 lines of third-party code are included to call five lines of one function. He even cites Halvar Flake, whose company cut organizations’ cloud bills by 50–90% and still failed because organizations weren’t interested — proof that the deeper fix is treating compute as a finite resource again. Cloud gave us the illusion of infinity (“click a button, add more RAM”), yet RAM and storage prices have risen 5–10x and even Google Workspace has capped its previously “unlimited” storage. Dependency reduction, he argues, follows naturally and shrinks the attack surface as a side effect.
Why Black Hat India, and Why Now
- Anant: India is simply too big a market to keep ignoring. Until now, associating with the Black Hat brand meant traveling to conferences on three different continents (plus one more in-region but still far). A home-ground event removes the time and cost barrier.
- Neelu: It’s a “win-win both ways” — a closer conference is more approachable for both submitting and attending. Indian researchers were already submitting strong work abroad; a regional stage gives them optimism of selection and a level playing field, while regional security leaders get access, creating a virtuous cycle: more innovation → more submissions → more presenting.
Advice for Briefings Submitters
- Neelu flags a structural gap: India has no focused academic institutions dedicated to cybersecurity, so even strong research (new vulnerabilities, protocol issues) stumbles at “the last mile” — how to document, package, and present the work is part of the research lifecycle, not an afterthought.
- Anant describes the three-part prep series run at other Black Hat editions — how to start research, how to submit, and how to present once selected — with recorded sessions and sample submissions linked from the submission page. His recurring reviewer pain: a submission that reads “interesting, but not interesting enough,” followed by a 50-page white paper that blows him away — a talk that would have been selected if the submission itself had carried the insight.
- His concrete test: “You have to come down from that high and submit something which an unknown person can read and understand” — hand the cold submission to a friend outside the research and ask what they actually understand from it.
What the Board Wants to See More Of
- A Pwn2Own winner from India — Anant’s stated aspiration: “I have not yet seen a Pwn2Own winner from India. I want to see that.”
- Hardware security — fewer people are working on it; he wants more.
- Quantum — “up and coming… going to be the next buzzword after AI,” chimed both board members.
- India-scale innovation — Neelu wants security innovation that works at scale, since mid-tier Indian companies serve populations larger than several countries combined, and within 5–10 years she wants original Indian security innovation shared with the world — not just following what China or the US does.
- India-specific stacks and scale-native problems — Anant wants research on stacks like UPI (now being exported to other countries), and on problems that only exist at Indian scale: race conditions that don’t need to be engineered because “on the day a result is declared, the exam server feels a DoS attack — not because someone is attacking, but because people are just accessing the resources.” Researchers often hesitate to submit such work abroad as “too India-centric”; a home conference gives it a stage.
One Thing Every CISO, Researcher, and Practitioner Should Know
- Neelu (speaking from her own Black Hat speaking experience): it is one of the best places for your work to be seen, surrounded by the best researchers in India and the region — and the third leg researchers forget is impact. She cites in-toto, which began as academic research and is now being adopted across supply chain security by nearly all major organizations: your research can benefit the entire community.
- Anant frames the conference as a concentration of silos: CISOs can scan the agenda’s keywords and direct engineers to relevant tracks (e.g., an AppSec talk on HTTP headers). His strongest recommendation for any conference, not just Black Hat: attend the Arsenal open-source tools demos first. Unlike briefings — which you may apply in 3–6 months — an Arsenal tool can be downloaded and used while it is being presented, with the author standing right there to fix bugs and take feature requests. Anant, a multiple-time Arsenal presenter, says he has gotten bugs fixed, features added, and lifelong friends out of it. For product scouts, the business hall beats 200 glorified vendor calls: watch five minutes, stay if it makes sense, and note that the smaller players may be working on something more interesting than the incumbents.
Key Takeaways
- AI’s defender value is real but narrow today: pattern recognition, correlation, and investigation augmentation — best realized by moving intelligence from non-deterministic LLMs into deterministic systems and keeping “rented cognition” to a minimum.
- Attackers adopted AI faster and with less debate: open-sourced malware (Shai-Hulud, Myasnikov’s releases) now forks and mutates quickly, and encrypted exfiltration is erasing defenders’ tracking visibility.
- AI-generated code is arriving faster than it can be quality-checked; meet developers where they work (IDE, CI), and bake deterministic + probabilistic controls into platforms.
- Inventory remains the unsolved foundation — SBOM’s negative sentiment shows that even a good inventory fails without a narrative for what it’s for. Reduce dependencies, reduce attack surface, and treat compute as finite again.
- Black Hat India (Bengaluru, October 27–30, 2026) lowers the barrier for Indian researchers whose scale-native work — UPI, exam-server-scale traffic — has struggled to land on global stages. The Call for Briefings and Call for Tools are open at blackhat-india.com; lead with the insight, and test the submission cold on someone outside the research.
