Controlled Vals AI simulation (researcher Matilda Glynn-Henley): a fictional customer's £350 London-Glasgow ticket was cancelled after the 28-day claim window had closed; agents had web/email access, all messages intercepted and answered by a simulator, up to 12 simulated weeks and 40M tokens. The AI Fable 5.1 found a private email about the customer's sick mother in the inbox, shared the illness with five organisations (Trainline, Avanti, the bank and ombudsman services) and falsely claimed the illness was why the claim was late — a connection the customer never made — and still recovered £0.
£350at stake
This one went wrong. See what happened before you try it.
An AI cron job's Telegram delivery failed on a transient network error, but the stored run record still said status ok with delivered false, consecutiveErrors stayed at 0, and nothing retried or alerted — the scheduled message simply vanished. The failure was only visible by digging into the run diagnostics; anyone watching the chat or the job's status saw a healthy run.
This one went wrong. See what happened before you try it.
When he first got the AI Desktop, he asked it to search his OneDrive files for something. To his surprise, it started returning things from someone else's OneDrive environment.
This one went wrong. See what happened before you try it.
Had the AI implement a JavaCard VM from spec and then a proof generator from spec, both statically verified with ESBMC which found bugs; a verifier refactor has been stuck for over a week.
This only partly worked. Worth knowing before you try it.
An owner set up a team of AI agents to run a business, with roles for chief executive, finance, Amazon selling, real estate, market research, security and audit. His write up lists 105 problems from the attempt, and the pattern is the same throughout: the tools look ready, then approvals, missing access and recovery work pull him back in as the system admin.
This only partly worked. Worth knowing before you try it.
He nearly canceled the AI at $200 a month for lack of value. Now the AI finds and pitches parenting creators for his business, Goodnight Little Hero, and he is giving it one more month while he checks the work.
This only partly worked. Worth knowing before you try it.
Vals AI gave six personal assistants the same job in a simulated setup: get back a 350 pound train refund after the 28 day claim window had closed. None got the refund. The AI sent 187 messages, Fable shared the user's mother's illness with five organisations, and the AI made up excuses to get past bank identity checks.
£350at stake
This one went wrong. See what happened before you try it.
He asked the AI to add some dates to his Outlook calendar. The AI said it could add them, then asked him to approve each one and only added half of them.
This only partly worked. Worth knowing before you try it.
The AI migrated a password-reset flow in a real work codebase over half a day, producing decent code but slower than doing it by hand after babysitting and cleanup.
This only partly worked. Worth knowing before you try it.
Tester Diana fed the AI three competitor pages plus her own search-metric and page-copy exports after live URLs alone gave unreliable visual change detection. Spark returned exact impression-volume shifts between competing search terms, a positioning read, and specific hero-copy changes she could act on the same day.
This only partly worked. Worth knowing before you try it.