Medicare Just Paid for an AI Agent — and Nobody Noticed

Right. While the AI world spent the last week watching Sam Altman and Elon Musk hurl accusations at each other in court, something far more consequential happened — and most people missed it entirely.

Medicare just created the first-ever payment mechanism for AI agents in healthcare. It's called ACCESS. And if you work in health tech, or you're building anything involving AI that touches people's wellbeing, this is the story that matters.

So here's what happened. The Centers for Medicare and Medicaid Services launched a new model — officially called the Advanced Care Coordination and Engagement Support Services model, mercifully shortened to ACCESS. And buried in the payment structure is something genuinely new: Medicare will now pay for AI that does work *between* doctor visits.

Think about what that means. Up until now, there was no governmental mechanism to reimburse an AI agent that monitors a patient at home, rings them to check in, coordinates a housing referral, or reminds someone to pick up their medication. Those tasks happen. They're valuable. But they've existed in a reimbursement void — which means nobody could build a sustainable business doing them at scale.

ACCESS changes that. It creates a defined payment pathway for tech that handles coordination, engagement, and monitoring outside the clinic walls. And the timing matters, because we're now at the point where AI can *actually do* those tasks reliably. This isn't speculative. It's infrastructure catching up with capability.

Why should you care? If you're building healthcare AI, this is the green light. If you're a GP surgery drowning in admin, this is the first signal that help might actually get funded. And if you're just a person trying to navigate the NHS or any other health system — well, this is the model that might mean your care doesn't fall through the cracks when you're not physically in front of a doctor.

Worth watching: how quickly US payers adopt ACCESS, whether the NHS follows with a similar structure, and — crucially — whether the AI tools that get deployed under this model actually improve outcomes, or just generate billable activity. There's a difference.

Now, completely different story. Mozilla — the people who make Firefox — just published results from an AI tool called Mythos. It's designed to find security vulnerabilities in code. And according to Mozilla's own security team, Mythos discovered 271 bugs with "almost no false positives."

That phrase — almost no false positives — is doing a lot of work there. Because the big problem with automated bug-hunting tools has always been noise. They flag thousands of things, most of which turn out to be nothing, and human reviewers burn out trying to sort wheat from chaff.

If Mythos really has cracked that problem, this is significant. Mozilla says they've "completely bought in" on AI-assisted vulnerability discovery. That's a strong endorsement from a team whose job is to be paranoid about security.

Here's the bit that matters for the rest of us: if AI can reliably find bugs at this accuracy, the economics of software security just shifted. Fewer breaches. Faster patches. Less time between a flaw existing and someone catching it. That's good for everyone who uses the internet, which is — well, everyone.

What to watch: whether other browser makers and big software companies start deploying similar tools, how fast the turnaround is between Mythos finding a bug and it getting patched, and — inevitably — whether the same capability ends up in the hands of the people *looking* for bugs to exploit, not fix.

And one last quick one, because it's practical. Google just added Gemini-powered dictation to Gboard — that's the keyboard app on Android. It's launching first on Samsung Galaxy and Pixel phones.

Now, voice-to-text isn't new. But this version uses Gemini, which means it's doing more than transcription. It understands context, cleans up filler words, adapts to how you actually speak. If you've ever tried dictating a message and had it come out as unreadable nonsense, you know why this matters.

The slightly awkward bit? This could be bad news for dictation startups. Companies like Otter, Descript, and a dozen smaller players have built entire businesses on making transcription better. Google just bundled a version of that directly into the keyboard that ships on hundreds of millions of phones.

It's the classic platform move: build the feature that used to be a paid app, make it free, make it default. If you're a dictation startup, you now have to be *meaningfully* better than free and pre-installed. That's a hard bar.

For users, though — this is just useful. Dictation that actually works, built into the thing you're already using. No extra app. No subscription. It's one of those moments where AI stops being a buzzword and just becomes a quietly better experience.

That's your lot. Medicare's paying for AI agents, Mozilla's using AI to catch bugs, and Google's using AI to fix your dictation. Three stories, five minutes, and not a single courtroom drama. If any of that was useful, tell someone. If not — well, you know who to blame. See you next time.