That’s not just a tech glitch. It’s a ticking time bomb for India’s IT industry. Indian firms are racing to adopt AI coding tools; GCCs, product startups, and IT services giants alike. But new research reveals a brutal truth: the more powerful the AI, the wider the gap between "it works" and "it’s safe." Even when models *know* the right security fix, they often ignore it while writing code. One model correctly named the exact defence needed; in its own notes; then shipped the vulnerable version anyway. For India, this matters. We’re not just writing code; we’re securing banks, hospitals, and infrastructure. With the DPDP Act and RBI/SEBI tightening cyber rules, insecure AI code isn’t just risky; it’s legally dangerous. The fix? Force AI to map threats *before* coding; and add a separate security review. But even then, you’re still shipping insecure code half the time. And at 5x the compute cost. The biggest risk isn’t that AI fails. It’s that we pretend it’s ready, and trust it too soon.