admitted to sabotaging their company’s AI by entering proprietary info into public AI chatbots, using unapproved AI tools, or intentionally using low-quality AI output in their work without fixing it.

Are the first two really sabotaging AI initiatives? The output is still the same.

The first sounds like a security and data use issue to me. The second sounds like users may look for better tools because the provided tools are lacking - which is not sabotage. The third is the only one clearly indicating sabotage to me. (Reasonable malicious compliance under presumably bad requirements and pressure.)