Back Office BeatSubscribe

Published by J9 Systems. What's new in business tech, and what it means for yours.

From the brief of Friday, October 9, 2026 · No. 2

From the daily brief · AI

Anthropic's new Claude usage policy takes effect Nov. 12; high-risk rules unchanged

Most shops can ignore it, but accounting firms, law practices and agencies using Claude for client work should spend 30 minutes checking two sections.

By Carter Josephson · · 2 min read

What happened

Anthropic published a new version of its Usage Policy on Oct. 8, 2026, taking effect Nov. 12. It mostly clarifies existing rules, and adds:

  • a consolidated section banning deceptive campaigns and fake accounts, political or commercial
  • tighter weapons and surveillance language
  • controls for Claude connected to hardware that takes autonomous physical actions
  • a ban on sustained abusive behavior toward its models

Its high-risk requirements are unchanged but rewritten. When Claude affects someone's health, legal rights, finances, livelihood or access to essential services, a qualified human must be able to review and change its recommendations, and the affected person must be told AI was used.

What it means for a 10-to-50-person business

For a dispatcher drafting customer emails or a bookkeeper summarising invoices, nothing changes.

The two sections that matter for firms our size are high-risk use and deceptive activity.

Take a 14-person accounting firm where 6 staff use Claude to draft cash-flow and tax-planning recommendations for clients. That touches client finances. A qualified accountant must review each draft and be able to change it, and clients must be told AI was involved. The rules are not new, but the rewritten section now lists which recommendations are covered. If the firm has no disclosure line in its engagement letter, that is the gap to close.

A small law practice drafting advice on clients' legal rights is in the same position. So is any shop using Claude to screen job applicants, since livelihood is on the list.

Agencies should note the new deceptive-activity section explicitly covers commercial work. That means using fake accounts or posts to amplify content, or hiding who is behind a message.

HVAC, plumbing, landscaping and cleaning shops using Claude for scheduling, quotes or emails can ignore this.

What to do
Ignore

No action is needed for routine Claude use, but if staff use it to draft client financial, legal or hiring recommendations, have one person read the rewritten High-risk Use Case Requirements section before Nov. 12 and confirm a qualified reviewer signs off and clients are told AI was used.

What it costs
What it costs
Read the rewritten high-risk section, 1 partner
0.5 hr (≈ $50 at $100/hr)
Add an AI-use disclosure line to engagement letters or templates
1 hr (≈ $100 at $100/hr)
Claude subscription price change
none announced in this update
Doing nothing
$0 for most shops; firms giving client financial or legal recommendations without human review and disclosure were already outside the policy
Sources
anthropic.com (primary)

Questions owners ask

Do I have to change how my team uses Claude by Nov. 12?

Only if you use it for work that affects someone's health, legal rights, finances, livelihood or access to essential services. Anthropic says those requirements have not changed; the section is rewritten to say more clearly which uses are covered.

What counts as a qualified human in the loop?

Anthropic describes it as someone with the authority to review and, if necessary, change Claude's recommendations. For an accounting firm that would be the accountant responsible for the client, not an intern forwarding the draft.

Does the hardware rule affect trades businesses?

Only if Claude is connected to equipment that takes autonomous physical actions and could cause injury. In that case a qualified operator must be able to watch and stop it, and the equipment must hold a safe state if Claude disconnects. Using Claude to write quotes or schedules is not covered.

Three to five of these, every weekday morning.

Related