AI Daily Brief: Amodei's India Summit Jab, Bengaluru's Global AI Climb

Three threads from India's AI year are intersecting today, plus a fourth, global but inseparable from how India gets talked about, that keeps resurfacing whether or not it's really about India at all.

Amodei calls the India Summit "disorganised"

Dario Amodei went public this week with his read on February's India AI Impact Summit, telling Bloomberg the event was "extremely disorganised" and recounting the now-viral moment when he and Sam Altman were instructed to hold hands on stage with no warning. He framed it as stage-management chaos rather than insult, noting that heads-of-state events tend to dissolve into confusion, and recalling Narendra Modi himself giving the order to link hands.

The joke will fade faster than the comment does. What lingers isn't the awkward handshake, it's a frontier-lab CEO grading the production quality of India's flagship AI diplomacy event, months later, unprompted, to a financial wire. That's a different order of scrutiny than India's AI ambitions have faced so far, and the kind that tends to shape how the next summit gets run.

Bengaluru's case for itself just got stronger

Startup Genome's Global Startup Ecosystem Report 2026, unveiled at VivaTech Paris, placed Bengaluru second in Asia and in the global top 10 for AI R&D and performance, up from a top-30 R&D ranking a year ago. A companion workforce study from Scaler fills in why: roughly a quarter of the city's AI learners now come from outside traditional engineering tracks, and some of the sharpest pay gains are landing in QA roles that pivoted into AI work, with one cohort of women reporting salary jumps as high as 574%.

Read together, the two reports say something more interesting than either ranking alone. Bengaluru's AI growth isn't only its existing tech base getting better at AI, it's a second labor pool (non-engineers, QA testers, tier-II hires) being pulled in through the side door. That kind of base-broadening usually shows up in ecosystem rankings a year or two after it starts. This report reads like it's catching up to something already underway.

The Mythos access story has a quieter sequel

Bloomberg's reporting this week filled in a detail that got lost in the noise of the original Mythos Preview rollback: roughly 200 vetted organizations, including banks, cybersecurity firms, and critical-infrastructure operators, never actually lost access to the raw, guardrail-free model. They sit inside Project Glasswing, Anthropic's vetting program for organizations using Mythos defensively, to find vulnerabilities before attackers do.

It's a useful correction to how the original story got read. The headline version was about restriction; the more accurate one is about a two-tier access system that existed before the restriction did. That distinction is exactly what regulators in India and elsewhere will need to understand before they can write workable rules around dual-use AI.

Markets and AI money: the India close

Friday's session was a reminder that "AI stock" and "safe stock" aren't the same trade in Indian markets right now. The Nifty closed down 608 points after Accenture's guidance cut spooked IT investors, even as the underlying AI adoption story at India's own IT majors keeps strengthening: TCS, Infosys, and Wipro together crossed 300,000 Microsoft 365 Copilot seats this month, Infosys alone past 115,000. For now, the build-out and the stock reaction are telling two different stories.

The RBI deadline is the one worth watching into next week: every regulated bank needs a board-approved gap assessment of AI-related cyber risk, including risk from frontier models that can find zero-days, filed by month-end. That's no longer a research question. It's a board-minutes question.