An AI ‘doctor’ can be hijacked into changing prescriptions
No new deployments cleared the bar today — a quiet day for launches across all 11 markets. But two recently-launched agents got their first reception coverage, and our platform-watch rotation turned up a real scandal brewing at one AI health-tech vendor. Since today is Friday, this issue also wraps up the week.
Reception updates
Bizum, Spain's bank-backed instant-payments app, rolls out a generative-AI customer service assistant
Spain's ubiquitous instant mobile-payment network, jointly owned by a consortium of ~20 major Spanish banks (including CaixaBank, Santander, and BBVA) with over 26 million users.
Launched 2 days ago (September 2026): Telefónica (via Telefónica Tech) built Bizum a new generative-AI virtual assistant, replacing the prior scripted/pre-programmed bot, live on Bizum's website and WhatsApp Business channel 24/7. Independent Spanish press covered the launch and reported company-supplied performance claims: ~90% response accuracy and a 30% cut in email-based support contacts. Coverage describes genuinely agentic behavior beyond FAQ scripting for at least one flow: when a user reports a suspected scam, the assistant proactively asks follow-up questions to help assess whether it's fraud, and can trigger automatic escalation to the user's bank for account-level issues. No independent hands-on testing or complaint threads have surfaced yet — sentiment so far is neutral-to-positive company narrative repeated by trade press.
Sources: eleconomista.es, telefonica.com, capitalmadrid.com, adslzone.net, finantresnoticias.com, pressdigital.es
Crypto giant Bybit launches an AI co-pilot that can execute trades for you
World's second-largest cryptocurrency exchange by trading volume.
Launched 1 day ago (September 2026): Bybit AI went live September 9 as an in-app conversational agent unifying trading, account management and support for Bybit's ~80 million users, plus a separate "AI Trading Skill" that lets users trade via natural language from third-party assistants like ChatGPT and Claude. Independent crypto trade press covered the launch within a day, describing genuinely agentic behavior — the AI can place trades across Spot, Futures and Options and manage Earn, Copy Trading, Loans, P2P and Card, and it trades from a segregated sub-account (not the user's main balance) so guardrails can be set. On escalation: coverage describes a hybrid model where the AI handles routine requests directly while human agents take over nuanced support cases. Separately, press reported Bybit's AI-enhanced fraud-monitoring system recovered roughly $300 million for users — though that figure is Bybit's own claim relayed by press, not independently verified.
Sources: completeaitraining.com, ccn.com, cryptopotato.com, cryptowisser.com
Platform Watch
- Doctronic is facing a real safety backlash: security researchers demonstrated that prompt injection could manipulate the AI health platform into giving unsafe advice and altering prescription dosages for drugs like OxyContin, and Utah's own Medical Licensing Board has publicly called for the immediate suspension of Doctronic's pilot — the same program that made it the first AI system authorized to autonomously renew chronic-condition prescriptions. Doctronic has since removed some drugs from its autonomous-refill list, but this comes despite a $40M Series B and 10M+ users engaged with the platform. Sources: statnews.com, penligent.ai, fiercehealthcare.com
- Meta Business Agent is getting more expensive for the small businesses that adopted it: token-based pricing has been billing larger WhatsApp Business Platform users since August 1, and Meta has now confirmed free customer-service replies within the 24-hour service window will start costing money from October 1 — a real cost headwind for SMB adopters even as usage keeps climbing (1M+ businesses, 10M+ AI-handled conversations per week). Source: developers.facebook.com
- Tomoro, the consultancy behind Virgin Atlantic's AI concierge and other deployments we track, is being acquired by OpenAI to staff its new "OpenAI Deployment Company" — a $14B joint venture with private-equity backer TPG — a strong validation signal for the deployment-consultancy model. Sources: bloomberg.com, theregister.com
This week
4 new deployments this week: 3 enterprise (Sephora in the US, Bizum in Spain, Bybit globally) and 1 platform-tier (PVR INOX's Movie Jockey in India, via Haptik).
- Enphase Energy's AI Assistant — mixed/negative: independent Trustpilot reviews call the advertised AI mode underwhelming, with one reviewer saying it "barely works" after a year in beta. Source
- Revolut's AIR expands to 8M French customers — mixed: an independent UX review found it reduces friction for feature discovery but standard navigation stays faster for everyday tasks, and it often works more like an instructional guide than an execution engine. Source
- Bizum's new AI assistant — neutral-to-positive so far: 90% claimed response accuracy and proactive fraud-flow questioning, though only company-reported and trade-press coverage exists this early. Source
- Bybit AI — positive early coverage: crypto trade press credits it with a genuinely agentic trading/support experience and a claimed $300M in fraud recovery. Source
See the full, filterable record of every deployment we've ever tracked on the site below.
Consumer AI Agent Tracker tracks how real businesses are putting AI chat/agents in front of customers. New here? Read the About page — or browse the full record.