What the evidence says

Deloitte's 2026 survey of 25,000 UK workers found that 63% had used generative AI for work. Among users, one in three did so without their employer's knowledge and half reported no formal training. Most use remained concentrated in familiar tasks such as drafting emails, finding information, and summarizing, rather than redesigning the work itself. These are self-reported UK survey results, not an estimate for every workforce or proof that any particular rollout succeeded.

The UK National Cyber Security Centre describes shadow AI as use outside approved systems and processes. Its September guidance argues that employees may turn to unapproved services when authorized tools or policies do not meet their work needs. It recommends understanding those needs, encouraging disclosure, and offering safer alternatives. The guidance identifies a risk and a response; it does not establish the effectiveness of a specific intervention.

A different kind of silence appears in Singapore. Reporting on September 17, CNA quoted a government minister urging employers to speak candidly about potential job impacts and retraining rather than treating a quiet workforce as agreement. This is public guidance about worker anxiety, not a measured finding that a particular employer's staff are resistant or consenting.

Change Collaboration's interpretation

The first silence can conceal workarounds; the second can conceal uncertainty. Both can make a dashboard look reassuring while the actual work and its human consequences remain poorly understood. More communication alone will not solve either problem if employees cannot see a safe path to raise a concern, use a suitable tool, or learn what a role transition would mean.

The current Trend Radar places human-AI work redesign first and AI trust, accountability and oversight second. Those are editorial rankings, not causal estimates. Context Radar brings the worker-voice issue into focus through news, an event agenda, and an executive account of employee-led experimentation. An agenda states what conveners intended to discuss, and the executive account is self-reported; neither verifies that the proposed practices work.

Practitioner Pulse points to a separate audience: two public conversations from experienced content-strategy and UX professionals considering an OCM career pivot. They asked how transferable skills relate to sponsorship, readiness, resistance, and sustained adoption, including whether an expensive credential is necessary. Their questions do not measure employer demand or certification value. They do, however, illustrate why listening and synthesis must connect to ownership of changed behavior rather than stop at discovery.

A practical test for change leaders

Treat silence as a prompt for better diagnosis, not as an adoption score. In a specific workflow, the change team can separate what is observed from what people believe it means, then bring employees and managers into the design of a credible response.

  • Ask where people already use AI, what need the approved path does not meet, and what information they fear sharing. Protect confidentiality and do not ask people to disclose sensitive data into an unsafe channel.
  • Map the task and decision changes, including who checks outputs, resolves exceptions, and can decline an unsuitable AI suggestion.
  • Explain honestly what is known, undecided, and contingent about job impacts and retraining. Give people a route to challenge the plan without treating quiet as consent.
  • Provide role-specific guidance and practice, then measure safe use, decision quality, workload, and outcomes alongside tool activity.
  • Return to the workforce after launch: show what changed in response to concerns and which questions remain open.

What not to infer

Hidden use does not mean every employee is careless, and a quiet meeting does not establish support. The UK survey, Singapore remarks, cyber guidance, and two OCM forum threads observe different populations and phenomena. Their convergence is a design question, not a pooled statistic: can the organization make safe use, candid transition, and meaningful worker voice visible enough to guide the next decision?