Attackers can now run capable AI with its safeguards removed on hardware they control. SofTx helps Canadian hospitals, imaging networks, clinics, and health-tech teams find and fix the gaps first, from PACS and HL7 interfaces to the AI you are deploying.
AI has shortened the time attackers need. It has not shortened the time a hospital needs to patch a system that is delivering care.
64–100%
Guardrails bypassed
How often simple techniques got GLM-5.3 to act on malicious attack orders in simulated tests, up from 0% when asked directly.
Anthropic, 2026US$6.64M
Average healthcare breach
The highest of any industry, for the 13th consecutive year.
IBM, 202672 hours
Ontario reporting window
In-scope hospitals must report critical cyber incidents to the provincial CISO within 72 hours of confirmation.
O. Reg. 51/26No. 1
Cybercrime threat
Ransomware is the top cybercrime threat facing Canada's critical infrastructure, including healthcare.
Canadian Centre for Cyber SecurityClinical environments have risks that general IT security programs often overlook.
Ontario's privacy commissioner documented an imaging clinic likely breached through a dormant admin account. The attackers deleted its backups, up to 550,000 patient records may have been affected, and the clinic temporarily closed.
Interface engines connect nearly every clinical system. Classic HL7 v2 feeds have no built-in authentication or encryption, so anyone who reaches that network segment can often read or alter messages.
Stale accounts, shared service logins, and always-on vendor VPNs are easy for automated tools to find. They often lead straight to clinical systems.
Scribes, chatbots, and retrieval tools add new paths to patient data, including prompt injection, over-broad record access, exposed model servers, and unapproved tools staff adopt themselves.
Winnipeg's Health Sciences Centre reported a 2026 ransomware incident affecting facility maintenance systems. Reports said doors and HVAC were affected, while patient care continued.
In 2023, ransomware at a shared IT provider forced five Ontario hospitals to shut down internal systems and delayed patient care. Your exposure includes your suppliers' exposure.
“Security has to work around the realities of patient care, or it won't get done.”
Why clinicians are part of every engagement
Security work from a team that also builds and integrates healthcare software.
We map internet-facing systems, PACS, interfaces, vendor connections, and AI tools, then rank what an automated attacker would reach first and what would hurt care most.
Authorized security testing of patient portals, clinical apps, APIs, FHIR services, DICOM endpoints, and the custom integrations generic scanners tend to miss.
Reviews of LLM and imaging AI deployments for prompt injection, record-level access control, model supply chain, logging, and unapproved AI use.
We are software engineers, so an assessment doesn't have to end with a PDF. We can patch the code, harden the integration, and verify that the fix holds.
Timed restores from offline backups, tested downtime procedures, and tabletop exercises with clinical and executive leaders before a real incident.
Practical support for cyber security maturity assessments, PHIPA safeguards, and vendor reviews, working alongside your privacy office and legal counsel.
Most of what limits the damage in a healthcare incident has to be in place before it starts.
Reactive
Proactive
Patches wait for the next maintenance window
Exposed systems follow a faster, risk-ranked patch schedule
Backups are assumed to work
Offline restores are tested and timed
Dormant accounts are found during the investigation
Dormant and vendor accounts are reviewed and removed regularly
The first downtime drill is the real incident
Clinical leaders have already rehearsed it
AI tools appear without a security review
Every model, endpoint, and data path is inventoried and tested
A repeatable cycle, because your systems and the threat will both keep changing.
Inventory the systems, data flows, suppliers, and AI tools that clinical care depends on.
Run authorized testing within agreed clinical safety limits, including AI-assisted attack paths.
Close the highest-risk gaps first, with engineering work your team can maintain.
Test restores and incident response, then retest as your systems and the threat change.
Security work should also leave you with the evidence regulators and boards expect.
In effect since July 1, 2026 for Group A, B and C public hospitals and the University of Ottawa Heart Institute: named contacts, maturity assessments every two years, and 72-hour reporting of critical incidents.
Health information custodians must protect personal health information and notify patients and regulators of certain breaches. Requirements vary by province, so we map yours first.
Private-sector organizations covered by PIPEDA, including many health-tech vendors, must report breaches that create a real risk of significant harm.
AI-driven threats, Canadian rules, and how we test safely in clinical environments.
Ask us something elseIt finds and reduces the risks specific to clinical environments: exposed PACS and DICOM services, HL7 and FHIR interfaces, vendor remote access, medical and facility devices, and AI systems handling patient data. SofTx adds software engineering to that work, so the code and integrations behind a finding actually get fixed.
Public threat reports from Google and Anthropic describe attackers using AI agents for reconnaissance, phishing, credential harvesting, and exploit development, often with people only choosing targets and reviewing results. In healthcare, the practical effect is less time between a public security fix and a working attack, while hospitals still patch around clinical schedules.
Anyone can download an open-weight model and run it privately, outside any provider's monitoring, and its safety training can be removed. NIST's CAISI called GLM-5.3 the most cyber-capable open-weight model released to date. Open weights are not the problem on their own, and defenders use them too, but hospitals should assume attackers have this capability.
Since July 1, 2026, Group A, B and C public hospitals and the University of Ottawa Heart Institute must designate cyber security contacts, complete a cyber security maturity assessment every two years and submit a summary, and report critical incidents to the provincial CISO within 72 hours of confirmation. We help teams prepare the evidence; your legal counsel should confirm how the rules apply to you.
It has to be planned that way. Scope, timing, and stop conditions are agreed with your clinical and IT leads before any testing starts. Intrusive tests run against non-production copies or in approved windows, and we do not run destructive tests against systems that are delivering care.
Yes. Self-hosting can be one of the strongest options for keeping patient data inside your network. It needs the same care as any clinical system: verified model sources, no open internet access, tightly scoped tools, patient-level access controls on retrieved records, and full logging.
Explore other solutions
Tell us what you run and what worries you most. We'll show you where to start.