Hackers weaponize AI agents as healthcare embraces them and EU tightens rules

Healthcare seems to have quietly become the proving ground for AI agents, and today's news makes that clearer than usual. MUSC Health is expanding its deployment of "Emily," SoundHound AI's agent, deeper into daily operations, while WellSpan Health has just signed a multi-year deal with Hippocratic AI to co-develop clinical agents — an extension of a relationship that started back in 2024. These aren't pilot programs anymore. When a health system commits to a multi-year partnership rather than a proof-of-concept, that's a signal they've moved past the "let's see if this works" phase into "this is now part of how we operate."

What strikes me about both deals is how little fanfare surrounds them compared to, say, a new frontier model launch. Nobody's calling Emily's expansion a breakthrough. But quietly integrating conversational AI into scheduling, triage, or patient communication at scale is arguably more consequential for most people's lives than another benchmark score. Healthcare is also one of the least forgiving environments for AI to operate in — the tolerance for hallucination or ambiguity is close to zero — so if these agents are sticking around and expanding rather than being quietly shelved, that tells you something about where the technology actually stands, not where the marketing says it stands.

Meanwhile, the enterprise AI agent market is dealing with a much less glamorous problem: nobody can agree on how to price the things. Some vendors charge per seat, some per task completed, some per outcome achieved, some bundle it all into existing SaaS contracts. I find this genuinely more interesting than it sounds, because pricing chaos is usually a symptom of a market that hasn't figured out what it's actually selling yet. Is an agent a tool, a labor replacement, or a service? Each framing implies a completely different pricing logic, and vendors seem to be hedging by trying all three simultaneously. For businesses trying to budget for this stuff, it's a mess — you genuinely cannot compare two competing agent products on price alone right now, because the units of comparison don't exist yet.

By the way, today also marks the day the EU's new AI content labeling rules take effect, requiring businesses to disclose AI-generated images, audio, video, and in some cases text used in advertising. This lands at an interesting moment — MiniMax just unveiled H3, a model generating 2K video with native stereo sound from mixed text, image, and video inputs. The generation quality is arriving faster than the regulatory and social infrastructure meant to handle it. I don't think labeling requirements alone solve the deeper trust problem, but they're at least an honest acknowledgment that the "was this made by a person" question isn't going away.

What happens when the agent pricing question and the healthcare deployment question collide — when hospitals start negotiating outcome-based pricing for clinical agents? That's probably closer than it sounds.