AI governance
Unmanaged AI gets expensive.
Or it gets banned.
Teams cap spend after bills spike. Agencies ban tools over data risk. See both happening below, and the one layer that heads off each without blocking your team.
Cost
The bill always arrives bigger than planned.
Lower token prices do not guarantee lower bills. More users, larger contexts, and agent workflows make spend harder to predict, and it's already forcing hard calls at companies far bigger than yours.
Uber · June 2026
$1,500 per tool, per month
Uber capped each employee's monthly spend on tools including Claude Code and Cursor after going over budget.
Read source · Bloomberg LawEnterprise survey · September 2026
1 in 3 exhaust token budgets early
In Accenture's survey of 750 executives, one in three organizations exhausted its token budget before year-end.
Read source · AccentureGitHub Copilot · June 2026
Model choice and tokens set the cost
GitHub moved Copilot to usage-based billing, with included credits and optional spending limits.
Read source · GitHub DocsDifferent companies, same pattern: usage outpaces whatever budget was set for it.
Sources reviewed October 23, 2026. Each link points to the publishing organization or the named news report.
Access
Public AI tools keep getting banned.
Four organizations below drew the same line in the past year: keep sensitive data out of public tools, and keep an approved route open instead of shutting AI down entirely.
Set clear access rules. Avoid blanket bans.
UK Department for Education · June 2026
Secure AI access remains available.
Departmental information cannot be used in public AI tools. Approved Copilot Chat and limited Claude access remain available in secure configurations.
Read primary source · UK ParliamentBritish Columbia Public Service · May 2025
The deployment and contract matter.
Confidential information is banned from public AI tools. Authenticated tools covered by the government's Microsoft agreement are allowed.
Read primary source · Government of British ColumbiaCenters for Medicare & Medicaid Services · Reviewed August 2025
Route by data sensitivity.
PII, PHI, and other sensitive agency data cannot be entered into public AI tools.
Read primary source · CMSUK Department for Transport · June 2026
Approved cloud AI is still an option.
Consumer AI outside the department's secure tenancy is not allowed for official work unless approved. Enterprise tools and specialist cloud access remain available.
Read primary source · UK ParliamentHow Nexus helps
One layer.
Every approved route.
The cost spikes and the access bans above aren't edge cases. They're the default now. Connect your apps once, and the same access, cost, and routing rules apply everywhere, automatically.
Apply policy. Choose a route.
One endpoint. Policy chooses the destination.
Route by sensitivity
When policy says sensitive work can't reach a public tool, Nexus sends it to local or private capacity automatically, so no one has to remember the rule.
Control access
Issue scoped app tokens instead of handing out provider credentials, so access can be revoked without touching every tool that uses it.
Apply guardrails
Catch PII, block risky keys, and rate-limit by app: the same checks agencies like CMS require before work ever reaches a model.
See consumption
Watch spend by app, model, and team as it happens, so a budget doesn't blow up before anyone notices.
A practical AI policy
Local for control.
Online when approved.
Choose the route based on the data, provider terms, model quality, and cost.
Explore Nexus EnterpriseTalk to us
How is your company governing AI?
Have similar policies? Need one place to manage access, routing, and spend? We'd like to hear what your team is working through.
Contact us