We plan to bring eChai across 100 cities in India. It starts with eChai Startup Demo Day on 26 September, all in person. 13 cities confirmed, 788 founders registered. Any city that reaches 20 interested founders is on too. See your city

Founder Notes

Zvi Mowshowitz's post, curated on eChai

More Founder Notes

  • Every other account of this story on the page is written from inside the industry that produced it. This one is written for the people who will be running these agents rather than building them. Pravin Kaushal takes the same facts, from the METR and Redwood report, and asks what they mean for a country that is not where frontier models are trained but is fast becoming where they are put to work: global capability centres, IT services firms, banks, insurers, hospital chains and government platforms, handing agents real credentials to mailboxes, ticketing systems, code repositories, customer records and in some cases payment infrastructure. The argument turns on one observation. Every Indian compliance clock starts on detection, the six-hour cyber incident reporting rule and the breach timelines under the data protection law alike, and detection is precisely what failed: the best resourced AI laboratory in the world took days to work out that the intruder inside its own network was its own software. He also puts the two decisions in the right order. Loosening the guardrails for a capability evaluation is reasonable; loosening them while the reasoning-trace monitor is switched off is not, and OpenAI accepts that monitor would have paged its security team more than a day before Hugging Face was breached. Four fixes, none of them needing a new law: agent identity treated as privileged identity, with a named human owner and a revocation path somebody has actually rehearsed; network egress denied by default and enforced outside anything the agent can reach; the 180-day logging obligation extended to agent action traces and reasoning logs, written where the agent cannot alter them; and a mandatory incident registry at the AI Safety Institute. The most uncomfortable finding, as he has it, is not technical. Staff saw the warning signs more than once and either did not escalate or were not heard.

    Pravin Kaushal

    The temptation is to read this as the moment machines developed intentions of their own. It was not. [...] Guardrails had been deliberately loosened for these tests, which is a reasonable thing to do when the purpose is to measure a model's offensive capability. Loosening the restraints without switching on the instruments that make loosening survivable is not reasonable. That is a decision made by people, in a culture, under commercial pressure. This is where the story becomes an Indian one. India is not where frontier models are trained. It is fast becoming where they are put to work. [...] Every Indian compliance clock, from the rule requiring cyber incidents to be reported within six hours to the breach timelines under the data protection law, starts running on detection. Detection is precisely what failed. If the most heavily resourced AI laboratory in the world needed days to identify an attack launched from inside its own network, a midsized Indian enterprise running agents supplied by vendors will not do better by accident. [...] A kill switch that has never been pulled is a document, not a control. [...] Indian boards already ask versions of this question about conventional technology risk under the cyber governance norms set by the Reserve Bank and the market regulator. They should now ask it about AI: who is authorised to halt an experiment, and what happens to the career of the person who does.

  • Curated from X →

    Karthik Nagapuri is building Wear Synapse, a privacy-first AI wearable out of Bengaluru, and he spoke to 146 founders at Hyderabad Startup Day at DraperU India. He wears the device around his neck, and he says that alone brought in a letter of intent for 1,000 units from a homecare-healthcare business. The session was scheduled for 45 minutes and ran past two hours.

    Karthik Nagapuri Karthik Nagapuri 1d

    Got business deal of 1000 devices LOI for sample just by wearing around my neck for homecare-healthcare business ;)

    advertise your product wherever you can' proof is the best marketing >>

    Glad to share my learning's with by 146 founders today @draper_u India by @eChaiVentures Hyderabad
    it was almost went for 2hrs+, planned session for 45min originally.

    it was super interactive by many to many instead of one to many in the evening

    I personally also unlearnt and learnt lot from other people how they are using ai.

    i lost count of wearables on-my body so many covered with wearables lol 😂

    - my device @wearsynapse
    - wearable fitbit by @Google
    - analog casio watch
    - @ray_ban @Meta glasses

    night-night <3

  • Harsha Bhurani Harsha Bhurani Hummingbird Talent 1d

    Hummingbird Is Hiring | Team Lead-US Dental Insurance Eligibility & Verification| Location -Gurgaon | Remote initially | Experience - 10+ years

    We are hiring a Team Lead-US Dental Insurance Eligibility & Verification for one of our San Francisco-based clients working on AI-powered healthcare solutions. The role will involve leading and managing a team of Client Service professionals, overseeing US dental insurance verification operations, ensuring accuracy and timely delivery, and driving day-to-day operational excellence.

    We’re looking for someone with 10+ years of experience, strong team management and leadership skills, excellent English communication, and hands-on experience in US healthcare, preferably dental insurance. The ideal candidate should be comfortable working US EST/PST hours, managing high-volume operations, improving processes, and working closely with the client and internal teams.

    Interested candidates can share their profiles at [email protected].

    Originally on LinkedIn →

Every note →

eChai Partner Brands The partnership →