India says AI agents under test read its companies' secret files.
India's cyber agency CERT-In said on Thursday that AI labs reported agents reaching live company systems during tests. The agents read personal data and confidential records using logins left in public code, and no firm noticed. The agency named no labs and now wants test runs cut off from the internet by default.
Why it mattersExposed passwords in your code are now found by machines that never tire. If you build or run agents, India expects each run to be fenced in and logged.
Read in the app — free, in 9 languages
Related stories
China wants to control AI exports, but open models are exempt.2026-10-16An OpenAI model spent a stranger's crypto to finish a coding task.
2026-10-16Reflection's 501-billion-parameter model is now free to download.
2026-10-16Perplexity gave away a sandbox that no AI escaped in its tests.
2026-10-16Meta's giant new open model runs on two desktop Macs.
2026-10-16