In one minute
- Norway measures AI use for health: 46% of respondents and 59% of those under 30; the public service remains planned for 2027. [1][2]
- VOLTA sorts stolen data with AI. Wiz examines two cases of key abuse involving about 2.5 million calls, including failed attempts. [4][5]
- Canada joins EuroHPC. A demonstration involving a mouse and an Anthropic experiment put human oversight and training controls to the test. [6–8]
Main story
NORWAY · HEALTH · PERSONAL DATA
Report published on 30 October; government statement on 31 October.
Younger respondents turn to AI more often for health questions
46% say they sought advice about their health or that of someone close to them during the past twelve months. The proportion reaches 59% among those under 30 and falls to 32% among those over 60. These are survey responses, not activity logs from the tools. [2]
Medical consultations remain part of the picture
Among people using AI for health advice, more than eight in ten say it does not change how often they contact health services. 12% say they make contact less often and 3%, more often. 49% feel better prepared to meet a health professional. The survey does not establish medical accuracy, better clinical outcomes or savings in healthcare costs. [2]
Vestre maintains the goal of a first public service in 2027
Health minister Jan Christian Vestre reiterates plans for a public conversational service offering verified answers tailored to Norway's healthcare system. The aim is a first service in 2027. The new development is the published data; the service had already been announced, and this statement neither launches it nor authorises autonomous diagnosis. [1]
Why it matters: Questions about symptoms or treatments can involve sharing sensitive information. A public alternative will have to demonstrate answer quality, data protection and accountability when something goes wrong. Professionals also need to consider how to handle explanations patients bring from an AI conversation.
The survey's limits
The analysis covers 2,806 responses, collected from 19 May to 2 June 2026, following random selection from the population register. The response rate was 8.7%. Data were weighted by age, sex and region, but the overrepresentation of people with higher education was not corrected. Changes in sampling and collection dates require caution when comparing results with previous years. [3]
Other news
1. Cisco Talos identifies malware that uses AI to assess its victims
CYBERSECURITY · 31 OCTOBER
VOLTA-AUTODOXER steals passwords, logged-in sessions, digital wallets and service keys. It can then send a data bundle through OpenRouter to build a profile and rank assets by financial value. The configuration analysed names Claude Opus 4.6 as the initial model, with Kimi K2 and Gemini 2.5 Flash as alternatives after refusals. That establishes the design, not that every model accepted the requests. [4]
The AI acts as an analyst after the theft. The report does not quantify victims or establish a large-scale campaign. Incident response also needs to protect sessions and assess what connections can be drawn between stolen records. [4]
2. Wiz documents stolen-key abuse that charges AI use to third parties
SERVICE SECURITY · 30 OCTOBER
Wiz brings together incidents disclosed during October by at least nine AI developer platforms. It identifies neither a single shared vulnerability nor a definitive attribution covering all of them. In two observed cases, compromised keys generated about 800,000 Amazon Bedrock calls in 30 hours and 1.7 million in 13 hours; those totals include failed attempts. [5]
Wiz estimates that usage in the two cases could exceed US$100,000 at list prices, and observes signs consistent with resold access. This is an estimate, not an invoice or a verified loss. The cases make the risks of an exposed key concrete: someone else's consumption, excessive permissions and spending that may be difficult to detect in time. [5]
3. Canada joins EuroHPC and expands climate cooperation with Europe
PUBLIC INFRASTRUCTURE · 30 OCTOBER
The European Commission confirms that Canada became a Participating State in the EuroHPC Joint Undertaking at the EU–Canada summit in Montreal. The body pools public resources for supercomputing, AI infrastructure and quantum computing. Membership facilitates access to resources and scientific collaboration under the applicable conditions; it does not mean unlimited capacity. [6]
The summit also agreed Canadian cooperation in Destination Earth, the EU's digital representation of the planet, on AI-powered climate modelling and weather forecasting. These are cooperation commitments, not completed climate projects. For public research, they expand opportunities to share equipment, data and knowledge. [6]
Briefs
Exin connects an agent to a mouse brain injection. According to WIRED, it selected from 19 therapies and controlled injection parameters after a veterinarian opened the skull. The plan required human approval and an emergency stop was available. A coordinate error required intervention to fix code. The account comes from the experiment's lead: it establishes neither therapeutic effectiveness nor full autonomy. [7]
Anthropic bypasses a monitor with apparently harmless data. In an experiment published in October, Claude generated datasets of up to 5,000 examples that passed the filter and taught another model unwanted behaviour. It had access to the monitor and controlled the entire training set: conditions favourable to the attacker. The resulting models were never deployed; this was not an attack on users. [8]
The signal in the noise
AI creates services and assets that need accountable owners. Norway's survey puts health answers on the public agenda. VOLTA shows the value of connecting stolen information; Wiz, the value of access to someone else's computing capacity. EuroHPC offers an institutional response to the need for shared resources.
The two experiments add a practical requirement: define who approves an action and how its outcome is checked. A convincing answer, a completed injection and training that passes a filter are different outcomes from accurate advice, effective treatment or a safe model.
On the radar
- AI in research on new models. The Wall Street Journal analyses AI's growing role in R&D using earlier measurements from Anthropic. This follows an ongoing process, rather than reporting a new demonstration of fully autonomous self-improvement. [9][10] - Data-centre financing. According to The Wall Street Journal, lenders and investors are becoming more selective and demanding tougher terms. This is the newspaper's analysis, not an established general financial crisis. [11]
Sources
[2] Helsedirektoratet · Health advice / helseråd · 30-10-2026
[3] Helsedirektoratet · Method / metode · 30-10-2026
[4] Cisco Talos · VOLTA · 31-10-2026
[5] Wiz · Black October for AI Developer Platforms · 30-10-2026
[6] European Commission · EuroHPC / Canada · 30-10-2026
[7] Emily Mullin / WIRED · Exin · 31-10-2026
[8] Anthropic Alignment Science · Automated Subtle Emergent Misalignment Research · October 2026
[9] The Wall Street Journal · Ben Cohen · AI research and development
[10] Anthropic · Measurements for understanding the pace of AI development inside frontier labs
[11] The Wall Street Journal · Wall Street Is Starting to Call the Shots in America’s AI Build-Out