Restricting access is an abstract idea; the mechanism that restricts it need not be.

The Federal Circuit has held that controlling access to resources, verifying a person's identity and filtering content by its characteristics are abstract ideas, and a claim that recites one of those goals carried out by generic computer components is ineligible. The same court has upheld claims to a behavior-based technique for identifying malicious code, and to storing a verification record in a particular region of memory because it is harder to tamper with there. The subject matter was security in every case; the difference was whether the claim recited how the protection is achieved or only that it is.

Trust scores and risk assessments are especially exposed, because a claim that gathers signals, computes a score and reports it follows the collect, analyze and display pattern the court has rejected repeatedly. The specification has to present the signals as technical ones — what a device, a connection or a storage system actually exhibits — and tie the score to something the system then does differently: refusing a session, quarantining a file, re-provisioning a device, isolating a node. Security is a property of the machine, and an improvement to it can be an improvement to the computer itself, but only when the claim reaches the machine.

Security inventions are usually built from parts that are individually old: hashing, signatures, key exchange, challenge and response, a second factor. An examiner who finds each part in a different reference needs little more than a reason to combine them, and "to improve security" is offered as that reason so routinely that it can appear to justify adding any safeguard to any system. What is new is the arrangement — what is bound to what, which party holds which key, the order in which the steps run — and the attack that arrangement closes.

That argument is strongest when it is in the specification before it is needed. An application that names the attack the earlier arrangement was open to — a replayed credential, a cloned device identity, a replication target that cannot be trusted with plaintext — and explains why the obvious safeguard does not stop it gives the response to an obviousness rejection something in the record to cite. One that describes only the arrangement leaves the reason it works to attorney argument, which is not evidence, or to a declaration assembled after the fact, whether before the examiner or years later before the Patent Trial and Appeal Board.