I attended the 2026 BASSCOM AI (Bulgarian Association of Software Companies) conference last week, quite an interesting event. One of the organizers and presenters, Moni Dochev, shared in his session that he wanted to machine-process the messages in the BASSCOM AI‘s Viber group, but failed to decode the now-encrypted Viber DB, so he hacked his way in by creating an automation that took screenshots of the tens of historical pages, then AI processed the content. Not ideal.
I took on the challenge and tasked Claude Code with it, purely for research purposes. Seven hours later, “it” succeeded – decrypting the local Viber database on my own Mac, containing only conversations I was a party to. No servers, accounts, or other people’s devices were touched. I am not going to share how “it” did it, nor the data it harvested out of ethical concerns, but the point is that AI agents are becoming dangerously good. The technical barrier has effectively collapsed – what’s left is intent. Anyone curious enough to ask is now capable enough to succeed.
