The Healthcare edition
This is a real edition from 27 October 2026, at the Practical level, in plain language.
Speech technology is coming into healthcare, while fake help desks, leaking share links and a PHP flaw show how quickly data can get out.
Sensire is building its own platform that turns conversations into reports
Care organisation Sensire is building its own platform for voice-driven reporting. It is called Spraak naar Waarde. The goal is less admin and more time for the client. There are plenty of programs that turn speech into text. But according to Sensire, they do not fit well with daily practice. By building its own, Sensire can take account of care terms, abbreviations and work processes. The first application is a Minutes app. An employee uploads a sound recording. The app turns it into minutes, with action points if wanted. The employee stays responsible for the report and adds to it where needed. Sensire is also working on applications for meetings with several care providers, behaviour consultations and conversations in which the care need is explored. Users test new applications first and give feedback. A wish for later is that information from conversations goes straight to the right place in the electronic client record. Until that link exists, that part of the work stays the same.
Speech to text is becoming an ordinary work tool. Sensire already uses it to make minutes, and Mistral is building agents that can also book an appointment. It saves time, but only if client data stays inside tools that your organisation has approved.
Fake help desk in Teams: attackers sometimes get in within fifteen minutes
Attackers pretend to be IT support in a Teams chat. Security company Arctic Wolf studied more than 400 such incidents. There is no fault in Teams itself. The attackers use external access, a function that is on by default. With it, someone from another Microsoft 365 environment can find, message and call employees. Both environments must allow this. It often starts with mail bombing. The victim's inbox fills up within minutes with hundreds of newsletters and confirmations. Shortly after, a helpful 'IT employee' makes contact via Teams. The goal is usually access to the computer through Quick Assist, a help tool in Windows. The victim only has to read out a six-digit code. If the victim cooperates, the harmful software is running within 15 minutes. Attackers increasingly choose a paid domain and clever names such as 'IT Support (Internal)'. Teams still shows the sender as external. Many care organisations use Teams, so such a chat could reach you too. Never read out your code to someone who sent you a chat themselves. Read more →
Meta Muse: shared conversations turn out to be findable
Meta Muse is a new general-purpose AI agent from Meta. Users can share conversations through a web link. Researchers at security company Claroc report that these links are easy to find. With one search they found 6,800 links. They opened a few hundred. These contained information that was not meant to be public, such as a business loan application, an internal revenue report, a holiday calendar for a whole company, notes for a confidential job interview and a bank fraud report with part of an account number. Meta says sharing is now paused and is considering changes to how findable the links are. Claroc reported this on 8 October through Meta's bug bounty programme. According to Claroc, there was no reply, not even to messages via LinkedIn. This is a general tool for consumers. Read more →
Microsoft AI makes the right diagnosis more often than doctors in a test
Microsoft has made an AI system for medical diagnoses, MAI-DxO. It imitates a team of doctors. Different AI roles discuss possible diagnoses, ask for more information, choose tests and check each other's conclusions. Researchers tested it in Science with 304 difficult cases from The New England Journal of Medicine and 325 files from an emergency department. The system had the right diagnosis in 83.2 percent of the difficult cases and in 80.3 percent of the emergency cases. General doctors scored 36.1 and 46.8 percent. The comparison has limits. In the difficult cases, the doctors were assessed on a smaller selection than the AI. They were also generalists, not specialists or teams. It was not studied whether cooperation between doctor and AI leads to better diagnoses or better outcomes for patients. The researchers therefore mainly see the system as extra support for diagnosis and in training. Microsoft now wants to test it with real patients at Beth Israel Deaconess Medical Center in Boston. Read more →
Mistral releases Voxtral 3 for voice agents
Mistral has released Voxtral 3. It is a group of models for speech to text and text to speech, with an SDK for building voice agents. Such an agent can listen, think, call a tool and talk in one continuous flow. Mistral mentions contact centres, service desks, virtual assistants and hands-free helpers. The transcription model works in more than 50 languages, recognises who says what and notices when someone has finished speaking. Mistral says a turn is done in less than a second. These are Mistral's own figures. Voxtral 3 is available through Mistral's API. Organisations can also run it on their own infrastructure. This way they decide where audio is processed and whether conversations are stored. The demo shows a patient with a French referral letter. The agent reads the letter aloud and books an appointment with a dermatologist. This is a demo, not described use in healthcare. Read more →
Google weakens privacy promise about Gmail and ads
Google changed its privacy policy on 16 October. Before, it said Google does not show personalised ads based on your content from Drive, Gmail or Photos. That firm promise has been weakened. Android Authority was the first to spot the change. According to Ars Technica, it concerns data you share through Personal Intelligence, a function from early 2026 that works with the Gemini app and AI Mode in search. The Gemini app currently has no ads. AI search functions can contain sponsored content. Google says linking is fully opt-in and that sensitive information is not used. According to Google, ads stay based on your search question, the context of the answer and your interests. Ars Technica reads it differently. If Personal Intelligence is on, summaries from your data can also count towards ads in AI search. This concerns the normal consumer version of Google. Read more →
Flaw in PHP knocks many websites offline
Many websites went offline as a precaution, including DigiD, MijnOverheid, RTL Nieuws and the website of the Tweede Kamer. The cause is a security flaw in PHP, a programming language that many websites run on. There are two flaws, both with a score of 9.2 on a scale of 0 to 10. Anyone who abuses such a flaw can break into a site and, for example, steal customer data. The makers of PHP first announced that a fix would come on Wednesday. Hackers used that announcement to track down the flaw. A fix only came around four o'clock in the afternoon, while the flaw already seemed to have been abused for hours. No concrete cases of abuse are known yet. According to NOS, there is a big chance that they will follow. A British security expert sees no signs of successful hacks. Cloudflare did see millions of attempts. Services also went down outside the Netherlands: 32 government services in the UK and some services at French company Orange. Read more →
Do you use Gmail, Drive or Photos privately? Then think carefully about whether to turn on Personal Intelligence, because that function gives Google AI insight into your mail and photos. Keep work email and client data out of your private account, so they do not end up in such a link.
AI and tech, made simple.
This is a sample of the Healthcare edition. Subscribers get it in their inbox every week.
Subscribe
inklaretaal · Amsterdam · © 2026
This is the generic Healthcare edition. Subscribe and you get it made to measure: tuned to your level, your job and the topics that matter to you. That way every week gets a little more personal.
Start for free