IBM Functional Testing

AI Security Review

Evidence context: threat category none; evidence quality moderate.

IBM Functional Testing generates Galasa functional tests inside VS Code and is published by IBM. The manifest lists no permissions and no host permissions, so the extension works through the ordinary VS Code API surface rather than any broad platform grant. Only one network endpoint in the list has the shape of a real address, api.dataplatform.cloud.ibm.com, IBM's Cloud Pak for Data API. A tool that generates and runs tests against IBM's platform needs to reach that host. The remaining entries are not endpoints: accountcreationtest.java and constants.java are Java source filenames parsed as hostnames, at.ch and al.bo are two-letter fragments lifted from minified strings, csslineclass.info is a CSS class name split on a dot, and collect.pro and cache.pro are property-access chains. That accounts for the bulk of the 342 indicator entries and tells you nothing about what the code does on the wire.

The eleven critical-severity hits all name one file family: extension/prebuilds/keyring.darwin-arm64.node, extension/prebuilds/keyring.win32-x64-msvc.node, and the rest of the platform builds under extension/prebuilds/. These are prebuilt native addons shipped by the keyring npm package, which stores secrets in the operating system keychain. The scanner labels them OBFUSCATION-NATIVE_BINARY_ADDON because compiled machine code is unreadable to a text-pattern engine. Every extension that ships a native addon produces this same finding. A .node file is not packed, encrypted, or self-modifying, and nothing in these eleven entries shows otherwise.

Credential access is where the verdict could have gone the other way. There are zero secret findings. Nothing reads .env, .ssh, .git/config, or cloud credential files. The only credential-adjacent component is the keyring addon, and it handles secrets the correct way, through the OS keychain binding, which is exactly what an IBM Cloud-authenticated test tool needs. Process and filesystem access are equally ordinary: test generation writes Java test files into the workspace and invokes the build. The malware and malware-signature categories are both empty, so no postinstall payload, dropper, or shell execution matched anywhere in the package.

The strongest counterargument is the combination of native keychain binaries and a live IBM Cloud endpoint. Those two together mean this extension can hold account credentials and reach the network, and if the publisher were unverified or the addons had been swapped, that pairing would justify a much harder look. IBM publishes the extension, the addons come from a widely used library, and no tampering indicator, unsigned binary flag, or outbound host outside IBM's domain shows up. The zero install count means there is no community track record to lean on yet, which is worth monitoring rather than treating as a signal of intent.

Key Reasons

• All 11 critical findings are OBFUSCATION-NATIVE_BINARY_ADDON hits on prebuilt keyring .node addons under extension/prebuilds/, which are machine code by definition and unreadable to a text scanner.
• Zero malware, malware-signature, secret, and tool-poisoning matches; no postinstall or dropper behavior anywhere in the package.
• The 342 network indicators are parsing artifacts such as accountcreationtest.java, constants.java, at.ch, al.bo, and csslineclass.info, while the only real host is api.dataplatform.cloud.ibm.com, IBM's own platform API.
• Published by IBM with no declared permissions or host permissions; the keyring addon is the standard OS keychain binding an IBM Cloud-authenticated tool needs.

False Positive Considerations

• OBFUSCATION-NATIVE_BINARY_ADDON firing on compiled keyring .node prebuilds for every platform
• IoC extractor reading Java filenames and minified-code fragments as hostnames
• Low-severity code-smell volume across bundled dependencies
• IoC count inflating severity despite containing no routable or suspicious host

Reviewed 2026-09-30; recommended action: no action; model confidence 86%.