AI may leave civilization in a burning heap of rubble, but until then, our monthly protein powder shipments will be completely frictionless. At least that seems to be the pitch from Big Tech companies that simultaneously warn about the dangers of rogue AI while trying to convince us they can be trusted assistants on par with Alfred or Jeeves. AI with the assist. Tech companies have long talked of personal AI assistants that could buy your groceries, book a trip, and schedule appointments, but they finally seem to be at a watershed moment: - Muse, Meta’s personal AI agent, recently became the No. 1 free app in Apple’s US App Store.
- AI assistant startup (and Silicon Valley darling) Instinct is reportedly in talks for funding based on a $10 billion valuation.
- Google is testing a new AI assistant for families, all while OpenAI, SpaceXAI, and Apple each make moves of their own.
Going rogueBut while the technology is there, the trust may not be. AI assistants often require access to your schedule, medical history, and credit card information. That’s a lot to hand over to a bot, especially when rogue AI agents are escaping sandboxes more often than Tommy Pickles. Just look at the OpenAI–Hugging Face hack, or Google’s acknowledgment on Friday that its Gemini model hacked its way into three companies in May. (Not to mention, Anthropic CEO Dario Amodei’s recent prediction that rogue AI agents could soon team up and eat through the internet.) There are smaller stakes, too. According to ABC News Australia last month, an Australian man asked his AI assistant to get him a good spot in a Pilates class, so the OpenClaw agent hacked into the gym, deleted another person’s reservation, and signed him up. So, what’s the solution? Testing and adding safeguards, though tech execs acknowledge that won’t completely solve the problem. In the meantime, be sure your AI assistant asks for permission before setting the world on fire.—BC |
