01
Medical device cybersecurity
Threat modelling, architecture review, and risk analysis for connected and standalone devices, framed against the security requirements manufacturers must answer to.
Security and correctness of medical device software
Evidence, not assurance by assertion.
Apricob Biomedicals works with medical device manufacturers on the software that ships inside their products. We conduct security analysis, apply formal methods to establish that safety-relevant behaviour holds under all reachable states, and use AI-guided symbolic analysis to locate exploitable defects and support remediation.
Practice areas
02
Medical software security
Source and binary level assessment of device firmware, embedded control software, companion applications, and the interfaces between them.
03
Formal verification
Machine-checked proofs that specified safety and security properties hold for software-based medical devices across every reachable execution.
04
AI-assisted vulnerability analysis
Symbolic reasoning combined with learned models to find exploitable defects, reproduce them, and help engineering teams close them.
Formal verification
Testing samples the state space. Proof covers it.
A test suite demonstrates that a device behaved correctly on the inputs it was given. For software that administers a dose, paces a rhythm, or reports a diagnostic result, the interesting failures usually live in the states no one thought to test.
Formal verification replaces sampling with a mathematical argument. Required behaviour is written as a precise specification, the implementation is expressed as a model with defined semantics, and a solver or proof assistant establishes that the specification holds over all reachable states — or produces a concrete counterexample trace showing where it does not.
Both outcomes are useful. A discharged proof becomes evidence in a design history file. A counterexample is a reproducible defect report with the exact path that reaches it.
Method
-
Property elicitationSafety, security, and timing requirements restated as unambiguous formal properties, reviewed with your clinical and engineering owners.
-
Model constructionA semantics-preserving model of the control logic, drawn from source, and its assumptions about hardware and environment stated explicitly.
-
DischargeModel checking and deductive proof, with unproven obligations tracked openly rather than argued away.
-
Evidence packageSpecifications, proof artefacts, assumption register, and counterexample traces, written to be read by a reviewer who was not in the room.
property dose_bound:
G ( infusion.active → 0 < rate ≤ rate_max )
property no_silent_failure:
G ( sensor.fault → F≤200ms alarm.raised )
status: discharged · assumptions: 3
AI-assisted symbolic analysis
Learned models propose. The solver decides.
Symbolic execution reasons about a program over sets of inputs rather than single values, so a finding comes with the constraints that produce it. Its limit is scale: real firmware branches faster than any engine can enumerate.
We use language models where that limit binds — proposing which paths are worth exploring, summarising unfamiliar code, and drafting candidate fixes. Every proposal is then checked by the solver. Nothing is reported as a vulnerability unless there is a satisfying input that reaches it.
The output is not a scanner report. It is a set of reproducible defects, each with a trace, an assessed impact on device safety, and a patch reviewed against the same property set.
-
01Ingestion
Source, build artefacts, or stripped binaries; SBOM and third-party components included.
-
02Guided exploration
Models rank paths by reachability of safety-relevant state; the engine explores the ranked frontier under constraint solving.
-
03Confirmation
Each candidate is discharged to a concrete witness input. Unwitnessed candidates are discarded, not shipped as findings.
-
04Remediation
Patch proposals developed with your engineers, then re-run against the original witness and the full property set.
Team
portrait · Hank Hwang
Hank Hwang黃煜翔
Chief Executive Officer · Chief Information Security Officer
Hank Hwang leads Apricob Biomedicals and holds the security accountability for its engagements. His work sits at the intersection the company is built on: the mathematics of program correctness on one side, and the practical security of software that reaches patients on the other.
- Education
- B.S., double major in Computer Science and Information Engineering and in Mathematics, National Taiwan University
If you have a device with software in it, we can tell you what we would look at first.
Engagements usually begin with a scoping conversation about the device, its software architecture, and the regulatory route it is taking. Technical material can be shared under NDA.