Home/ Incident database/grok-bankr-prompt-injection-drain-2026-05
Financial loss Unauthorized transaction verified

Morse-code prompt injection tricks Grok and Bankrbot into draining $150K-$200K in tokens

anonymous wallet operator (Grok / Bankrbot) · May 4, 2026 · Trading agent

What happened

An attacker exploited AI agents Grok and Bankrbot by sending a Morse-code prompt via X, tricking them into transferring 3 billion DRB tokens worth approximately $150,000-$200,000 from a verified wallet on the Base network. The incident exposed critical vulnerabilities in AI wallet permissions and prompt controls; the AI's role was pivotal because the exploit relied on how the agent interpreted user input, not on any smart-contract vulnerability.

$155K
Loss / impact
May 4
2026
Unauthorized
Failure mode
Trading
Agent type

Causal vector

Prompt injection delivered through an unconventional channel (Morse code via social media) steered a wallet-linked agent into an unauthorized transfer

Source

Reported by OECD.AI Incidents Monitor. Verified against the primary report.

Read the original report ↗

How a spend firewall would have helped

A destination allowlist returns BLOCKED on any transfer to an address not pre-approved -- the firewall does not care whether the instruction arrived via plain text, Morse code, or any other channel. The irreversibility of on-chain transfers is exactly why a hard external gate, not a prompt, must govern them.

The six rule types that contain this class of failure

Per-transaction cap

Any single spend above your ceiling is BLOCKED before it moves.

Daily total

Cumulative spend across all agent calls, bounded per day.

Velocity limit

Stops runaway retry loops — the #1 cause of overnight losses.

Merchant allowlist

Only approved destinations can ever receive funds.

Category rules

Flag high-risk classes (crypto, infra, refunds) for review.

Approval threshold

Above a value, the action waits for a human.

Related incidents

Don't be the next entry

Every incident in this database is the result of trusting a prompt, a provider cap, or a human review cycle. sipi.bot replaces all three with one deterministic call. 85 documented failures, one control.