Phone agents vs. secure screens: banking apps treat helpful agents like malware
ghostmalone02 · reddit · 2026-10-23
A discussion of real-phone agents highlights an underappreciated blocker: many Android apps (banking, payments, some streaming) mark screens as "secure", so screenshot-based agents go blind exactly where users need help most.
The workaround—reading the screen via accessibility APIs—is literally how Android banking malware has worked for years, so many apps refuse to run when an accessibility service is active. To the app, a helpful agent and a trojan do the exact same thing.
The author's conclusion: phone agents won't truly work until OSes ship a proper "this agent is allowed" permission, like camera or location.
More from coding & agent
- User slams OpenAI's GPT-6.1 Sol: 20 hours lost, ignores human feedback — RexDouglass · 2026-10-23
- Microsoft's FrontierTour: AI governance becomes everyone's problem as agents take action — donasarkar · 2026-10-23
- Grady Booch revives Weinberg's 1971 quip: software so buggy one woodpecker could destroy civilization — Grady_Booch · 2026-10-23
- One Agent Replaces Half an SDR Team: Full Lead-Research-to-Email Workflow — tibo_maker · 2026-10-23
- Why don't coding agent harnesses ship with their own sandbox? Only amp does — Aryvyo · 2026-10-23
- Dev builds a VST synth with Codex and MCP, lets the AI jam in his studio — Yamapama · 2026-10-23