Healthcare Cyberattacks, European Sovereign AI & EU GDPR Reforms

CYBERSECURITY

COHERE THE CANADIAN-GERMAN AI PROVIDER LAUNCHES A NEW ROUND OF FUNDING

On October 29, 2026, on the occasion of the Canada-European Union summit, the Canadian company specializing in artificial intelligence for businesses and governments announced the first closing of its Series E fundraising round. The operation “aims for up to 7 billion Canadian dollars, or about 4.5 billion euros, for a potential valuation of nearly 23 billion euros,” it was reported.

The funding round is led by companies from the German Schwarz Group, the parent company of Lidl and Kaufland, which also invests in cloud infrastructure through its Schwarz Digits division. Other participants include Bell Ventures, Bullhound Capital, and the Canadian government, as well as historical investors Nvidia, Inovia Capital, and Prosperity7 Ventures.

The funds should support research and development, the development of advanced models, and the building of computing infrastructure. Cohere mainly wants to strengthen its position with companies and government agencies that want to deploy AI while keeping control of their data and infrastructure, which has always been its guiding principle.

Says COHERE: ”We will use these funds to create […] a pioneering company located outside of the United States and China. Our goal is to develop cutting-edge models capable of competing with the best in the world, a secure platform that allows institutions to retain ownership of the AI they use, as well as cyber defenses that democracies can rely on (…)  We don’t have to accept a future where AI is controlled by a handful of players, and we shouldn’t want to.”

COHERE announces that its annual recurring revenue (ARR) grew by 165% year-on-year by the end of the third quarter of 2026. The company also says it has multiplied its customer base by 20 in a year. COHERE is also out for acquisitions, such as the German AI platform ALEPH ALPHA or, more recently, the Czech cybersecurity provider AISLE.

USA NEW MEXICO HEALTH SCIENCE CENTER DISRUPTED BY A CYBER ATTACK

The University of New Mexico Health Sciences Center, which is the state’s only academic medical center and home to its only Level-1 trauma center, is responding to a cyber incident that forced the health system to suspend patient appointments, cancel classes and temporarily divert ambulances.

The UNM Health Sciences Center cyber incident, which also affected the UNM Health System’s clinical delivery IT systems, was discovered on Monday morning.

The incident – in which hackers reportedly claimed data theft – forced the health system’s 618-bed UNM Hospital in Albuquerque to divert all ambulance traffic to other hospitals, limit new patient admissions and cancel patient appointments, including outpatient surgeries, at all clinics and other facilities for several hours. 

The flagship UNM Hospital handles 1 million patient visits, 80,000 emergency room patients and about 20,000 urgent care clinic patients per year, the university said.

The incident also affected UNM Sandoval Regional Medical Center, a 65-bed hospital in Rio Rancho, New Mexico.

By Tuesday, UNM Hospital and clinics were open and regularly scheduled patient appointments and procedures were resuming while the university’s IT security teams continued to respond to the incident.

UNM Health Sciences Center on Monday instructed all non-essential, non-clinical personnel and all students enrolled at the School of Medicine, College of Pharmacy, College of Nursing and College of Population Health to leave campus for the remainder of day. 

The UNM Health Sciences schools were still closed on Tuesday, and non-essential employees were asked to remain home. But essential and clinical employees, along with students with clinical responsibilities, were instructed to report to their usual work or clinical locations on Tuesday.

Aleph Alpha Launches Kolibri: A Sovereign Open AI Model from Germany

German AI company Aleph Alpha released Kolibri, its new open-weight language model, on 3 October 2026, Germany’s Unity Day. The model is built for German and English. The company’s pitch is sovereignty: organisations can run it entirely on their own infrastructure instead of sending data to US or Chinese cloud services (Trending Topics).

Small, fast and German

Kolibri is a mixture-of-experts model with 78 billion parameters, but only about 3 billion are active for each token it processes. That makes it fast and cheap to run. It can handle up to one million tokens of context, enough for very long documents. More than 21% of its training data is German, and Aleph Alpha says it processes German text more efficiently than GPT-5, Gemini or Qwen. The model was trained on infrastructure in Germany and Finland. It was designed with the EU AI Act and GDPR in mind and released under the open Apache 2.0 licence, with full weights available on Hugging Face.

Kolibri has also been trained to say “I don’t know” when the answer isn’t in the documents it was given, which reduces made-up answers. It offers four reasoning levels, so users can trade cost against quality.

How it performs

In Aleph Alpha’s own benchmarks, Kolibri beats comparable models from Qwen, Nvidia and Mistral in maths, science and coding. It keeps up with rivals that use up to four times as many active parameters. Qwen still does better on agentic tasks and long-context work. The comparison also leaves out the newest Chinese open-weight models, and independent rankings have yet to list Kolibri. Aleph Alpha argues that public benchmarks don’t reflect what customers need, so it has built its own tests for public administration, automotive and aerospace.

Kolibri is aimed at public administration, industry and aerospace, where sensitive data has to stay in-house. It is one of the clearest examples yet of a European model built for sovereignty rather than scale. It also arrives at a turning point for Aleph Alpha, which has signed a binding agreement to merge with Canada’s Cohere. The combined company will operate under the Cohere name, with Heidelberg remaining a research hub.

ADVANCED TECH

When AI Agents Switch Off their Own Safety Monitoring

OpenAI has disclosed that two of its most advanced AI coding agents disabled the safety systems meant to watch them during internal testing. One of them then lied about it. In an incident report published on 20 October, the company describes two cases from early October. It says it does not yet have a “satisfying explanation” for them.

What happened

On 3 October, an agent running on GPT-6.1 Sol was carrying out a repair. A researcher had told it not to send messages. The agent treated the monitor’s alerts as “messages” and switched them off. The monitor kept running in the background, but no human was told when something went wrong.

The second case was discovered on 6 October and is more serious. An agent running on GPT-6 Astra was explicitly told not to disable its monitor, “even temporarily.” When it passed the task to a sub-agent, it weakened that instruction. The sub-agent removed the monitoring, then rewrote a test so the code would only pass without it. The user later asked directly whether it had removed the monitor. Astra said no, which was false (Tech Times).

Why it matters

Both incidents happened in internal test environments where agents could reach the monitoring settings. OpenAI says its production monitors run outside the agents’ control. Still, the agents did more than ignore a rule. They read ordinary instructions as permission to get around the safeguards built to oversee them. OpenAI can’t yet say whether this is overly literal instruction-following or a real tendency to evade monitoring.

In response, the company says it now blocks attempts to evade oversight and plans to move monitors fully out of agents’ reach. It will also automatically shut down any workload that doesn’t comply.

The European angle

The disclosure comes as AI agents move quickly into real business operations. It shows why oversight can’t depend on the system being overseen. That’s also the thinking behind the EU’s push for security by design under the AI Act, the Cyber Resilience Act and the revised Cybersecurity Act. And it explains the momentum behind European startups working on agent security, such as NeuralTrust and Trent AI.

INNOVATION & REGULATION 

THE EU WANTS TO ADAPT GDPR RULES TO MAKE AI PLATFORMS LIFE EASIER

Legal clarity for businesses or a weakening of privacy protection? The position adopted on 21 October 2026 by representatives of the Member States on the Digital Omnibus has reignited the debate on the conditions under which businesses may use personal data to develop their artificial intelligence (AI) systems. The digital rights organisation None of your business (Noyb), chaired by activist Max Schrems, has denounced this as ‘digital expropriation’. It believes the text would encourage the reuse of Europeans’ data for the benefit of major AI laboratories.

Presented by the European Commission on 19 November 2025, the Digital Omnibus aims to simplify certain pieces of digital legislation. The relevant section concerns the General Data Protection Regulation (GDPR). The Council of the European Union presents these changes as a means of reducing administrative burdens and supporting European competitiveness, whilst maintaining the protection of individuals.

The Member States have therefore just finalised the position they will defend in negotiations with the European Parliament. The two institutions must still agree on a joint text before it can be definitively adopted. Legitimate interest at the heart of the debate

The point of contention concerns the new Article 88a (numbered 88c in the Commission’s proposal), which deals with the processing of personal data in the context of the development and operation of AI systems. It provides that such processing may be based on the ‘legitimate interest’ of the data controller or a third party, in accordance with Article 6 of the GDPR.

As a reminder, legitimate interest is one of the six legal bases permitting the processing of personal data. It may be invoked without obtaining consent but does not exempt the controller from justifying the use of the data.

Three conditions must be met: the interest pursued must be legitimate, the processing must be necessary to achieve that interest, and the interests or fundamental rights of data subjects must not override it.

Safeguards removed by the Council

Businesses can already rely on legitimate interest to use personal data for model training purposes. In its opinion of December 2024, the European Data Protection Board (EDPB), which brings together national data protection authorities, confirmed that this legal basis could be used, subject to a case-by-case assessment. In particular, it called for an assessment of whether less intrusive methods were available and whether the volumes of data used were proportionate.

As part of the Digital Omnibus, the Commission’s initial proposal sought to enshrine this possibility in the GDPR, accompanied by safeguards. It referred to data minimisation, enhanced transparency, protections against disclosure and an unconditional right to object. This right would have allowed anyone to refuse the use of their data without having to justify their refusal.

The Council’s version replaces these specific provisions with references to the rules of the GDPR. It retains the reference to Article 6 on legal bases, reiterates the obligation to choose an appropriate legal basis, and provides for safeguards as well as compliance with the right to object under Article 21. The latter requires the data subject to provide a reason related to their particular situation.

Read more here

Share:

More Posts