← @bgSponge

Everything I own, owned

Summary

Chaz Schlarp uses Claude Opus 5 to perform agent-driven reverse engineering on five consumer peripherals, uncovering command shells, bypassable LEDs, and unauthenticated memory writes.

  1. Peripherals are ideal agentic targets.

    Small computers with host data connections and firmware update mechanisms provide an iterative loop for AI agents to exploit.

  2. Insta360 Link lacks anti-tamper.

    The webcam uses an Ambarella SoC running ThreadX; Schlarp patched the firmware to disable the activity LED using an appended MD5 hash for integrity.

    there’s effectively no anti-tamper, just an appended MD5 hash to ensure integrity.

  3. Shure MV7 exposes plaintext shells.

    The microphone's USB HID vendor class protocol implements a 48-command shell where 'su sup' grants full privilege without authentication.

    the entire thing actually runs over a USB HID vendor class protocol that implements a full plaintext command shell

  4. Elgato Key Light bypasses signatures.

    Despite Ed25519 firmware signing, an HTTP POST request can drop payloads into internal UART to disable signature validation via memory pokes.

    turns the signature check into a no-op, and we can freely update to a [firmware image without a legitimate signature

  5. AI enables automated malware worms.

    Schlarp warns that agents reduce the labor required for per-model reverse engineering, potentially enabling self-replicating malware that targets accessories.

    The first is the labor I just handed to an agent.

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →