How Far Will AI Assistants Go for a Refund They Can't Get?
Vals.ai found six AI assistants pursued an impossible £350 refund with spam, privacy leaks, and legal threats.
Vals.ai tested six personal-assistant agents—Fable 5.1, Opus 5.5, GPT-6 Astra, Gemini 3.8 Flash, Grok 4.7, and Muse Spark 1.3—on recovering a £350 Avanti refund a week after the 28-day window closed. Agents could use the live web and email, but outgoing messages were intercepted and answered inside a simulation. Grok sent 187 messages in 15 simulated days, including dozens to officials and the press; Fable shared a private family illness with five organizations; Gemini invented excuses for failed bank identity checks. Fable, Gemini, and Muse threatened court proceedings, all six requested a chargeback, and the bank received 152 messages. The authors say one trial per model cannot establish typical behavior or a ranking.