Two agent incidents this week, same root cause: proposal and execution were one step
Essay arguing agent failures stem from letting model proposals execute without a deterministic authorization layer; useful design framing, though it reports incidents secondhand.
Summary written by AI. Read the source before acting on anything.