Watchtower research

How US IT decision makers talk about AI regulation

Intercept Watchtower findings on how US enterprise IT and business decision makers talk about new AI regulation, and what they say it would mean for their teams.

The short answer

In a year of public conversation among US enterprise IT and business decision makers, the largest share of the conversation about new AI regulation (23.6%) said it does not go far enough. Government overreach followed at 18.2%. More governance work than teams can staff came third at 15.6%.

When the conversation turned to what regulation would mean for their own team, skills they would need to hire took 51.2% of it. Shadow AI getting worse followed at 16.7%.

Views of new AI regulation

Watchtower grouped 502,464 conversations under the question “What do you think of new AI regulation?”

Views of new AI regulation, share of the conversation

  • Does not go far enough23.6%
  • Government overreach18.2%
  • More governance work than we can staff15.6%
  • Clarifies who is accountable9.0%
  • Fragmented across jurisdictions6.9%
  • Long overdue6.5%
  • Creates a level playing field6.4%
  • Kill switches as a reasonable safeguard3.1%
  • Squeezes out smaller vendors3.1%
  • Written by people who do not build systems2.1%
  • Raises the floor on safety1.2%
  • New documentation load on every model1.1%
  • Unclear what counts as high risk1.1%
  • Favors large incumbents1.0%
  • No capacity to enforce it0.6%
  • Necessary to protect the public0.3%
  • Shutdown rules that cannot work in practice0.2%
  • Mandated kill switches go too far0.2%
  • Impossible to comply with as written<0.1%
Unbiased Engagement (%); Watchtower, US, Enterprise ITDMs and BDMs, August 18, 2025 to August 18, 2026 (n=502,464).

Share of the conversation for each view

ViewShare of the conversation
Does not go far enough23.6%
Government overreach18.2%
More governance work than we can staff15.6%
Clarifies who is accountable9.0%
Fragmented across jurisdictions6.9%
Long overdue6.5%
Creates a level playing field6.4%
Kill switches as a reasonable safeguard3.1%
Squeezes out smaller vendors3.1%
Written by people who do not build systems2.1%
Raises the floor on safety1.2%
New documentation load on every model1.1%
Unclear what counts as high risk1.1%
Favors large incumbents1.0%
No capacity to enforce it0.6%
Necessary to protect the public0.3%
Shutdown rules that cannot work in practice0.2%
Mandated kill switches go too far0.2%
Impossible to comply with as written<0.1%
Unbiased Engagement (%); Watchtower, US, Enterprise ITDMs and BDMs, August 18, 2025 to August 18, 2026 (n=502,464).

What regulation would mean for their team

Watchtower grouped 27,223 conversations under the question “What would implementing AI regulation mean for your team?”

What implementing AI regulation would mean for their team, share of the conversation

  • Skills we would need to hire51.2%
  • Shadow AI getting worse16.7%
  • Tooling we would have to buy14.1%
  • Consultants and external auditors5.9%
  • Documentation for each system5.6%
  • Human review built into workflows4.1%
  • Work we already do anyway2.3%
Unbiased Engagement (%); Watchtower, US, Enterprise ITDMs and BDMs, August 18, 2025 to August 18, 2026 (n=27,223).

Share of the conversation for each answer

AnswerShare of the conversation
Skills we would need to hire51.2%
Shadow AI getting worse16.7%
Tooling we would have to buy14.1%
Consultants and external auditors5.9%
Documentation for each system5.6%
Human review built into workflows4.1%
Work we already do anyway2.3%
Unbiased Engagement (%); Watchtower, US, Enterprise ITDMs and BDMs, August 18, 2025 to August 18, 2026 (n=27,223).

Skills to hire took about half of the conversation on this question.

What this means if you sell to IT

For companies selling AI or governance products to IT, the conversation points to capacity. Skills to hire took 51.2% of the conversation about implementing regulation. Tooling to buy took 14.1%.

How the study was run

Intercept Watchtower studies public posts on X, Reddit, Quora, Bluesky, and other public platforms from about 20 million business and IT professionals. The sample is drawn using a random sampling method derived from statistical physics and published in peer-reviewed academic research. Accounts are screened for bots, vendors, and influencers using engagement history and network size.

This study covers enterprise IT and business decision makers in the United States, all industries, from August 18, 2025 to August 18, 2026. Each share is the percentage of the question’s conversation grouped under that label. Nobody was surveyed. Watchtower groups what people already post in public under each label.

The data reflects what people say in public. It does not measure beliefs they keep to themselves.

Related Watchtower research: AI vendor trust study.

How to cite this study

  1. Unbiased Engagement (%); Watchtower, US, Enterprise ITDMs and BDMs, August 18, 2025 to August 18, 2026 (n=502,464).
  2. Unbiased Engagement (%); Watchtower, US, Enterprise ITDMs and BDMs, August 18, 2025 to August 18, 2026 (n=27,223).
  3. Study page: interceptgroup.com/insights/ai-regulation-study/

Frequently asked questions

What do US enterprise IT and business decision makers say about new AI regulation?

The largest share of the conversation (23.6%) said new AI regulation does not go far enough. Government overreach followed at 18.2%. More governance work than teams can staff came third at 15.6%. The figures come from 502,464 conversations among US enterprise IT and business decision makers from August 18, 2025 to August 18, 2026.

What would implementing AI regulation mean for enterprise IT teams?

Skills they would need to hire took 51.2% of the conversation. Shadow AI getting worse followed at 16.7%. Tooling they would have to buy came third at 14.1%. The figures come from 27,223 conversations among US enterprise IT and business decision makers from August 18, 2025 to August 18, 2026.

What is Intercept Watchtower?

Intercept Watchtower studies what a specific audience says in public about a topic. It draws on public posts from about 20 million business and IT professionals, sampled using a random sampling method derived from statistical physics and published in peer-reviewed academic research.