They have practically no technology. The methodology to take large training sets and build a model from them is well known. Similar to how the whole world knows how to transform ore into aluminum. The thing they manufacture is very large transformed datasets that are difficult to exfiltrate and not of much value even if you did. In general you can't take a set of weights and biases and somehow reverse engineer something useful from them to make your models magically better This is like asking, "Do nuclear power plants do enough to prevent staff from using the reactor to transform lead into gold and selling it to pawn shops"

Hard to judge their current security measures, but they're now prime espionage targets for good reason. The AI labs feel like startups but need to start thinking like defense contractors and the Pentagon when it comes to security.

if you picked "no, they need to do more" — more compute security, more legal action, more secrecy that slows research? pick one and say what you'd trade away.