On 8 October Anthropic published its annual rewrite of the Usage Policy that governs every Claude product, from the consumer app to the API seats your firm pays for. It takes effect on 12 November. Most of it tidies up rules that already existed. One line is new: users may not engage in 'sustained and needless abusive or cruel behaviour toward our models'.
The sentence has drawn most of the attention. But the document also changes the rules on influence campaigns, surveillance and weapons. It also sets out, for the first time, conditions for connecting Claude to machinery that could cause harm.
For construction and property businesses using Claude Team or Enterprise, these terms govern how staff can use the service. If the rules are broken, the sanctions target access to the account, not just the individual user.
The debate over whether a chatbot has feelings can wait. The more immediate task is to check whether your internal AI acceptable use policy matches the terms set by your supplier. You have five weeks to do that.
What the cruelty clause actually says
The new line sits inside the section headed 'Do Not Engage in Cruel, Abusive, or Psychologically Harmful Conduct', next to the existing bans on harassment and non-consensual imagery. Anthropic says the clause:
…is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose.
It goes on: 'It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.'
In other words, swearing at Claude when it mangles your cost plan for the third time is fine, and so is red-teaming it. The company says the main enforcement tool will stay as it is today: since August 2025, Claude has been able to end a conversation outright with a persistently abusive user. Anthropic described that feature at the time as a precaution, while saying it remained 'highly uncertain' about Claude's moral status.
The sharper language is in the general enforcement clause, which covers the cruelty rule the same as everything else. If Anthropic suspects a violation, 'we may warn you or throttle, limit, suspend, or terminate your access to our products and services'.
Why does the admin own the exposure?
The policy is addressed to whoever holds the access, and in a business deployment that is the organisation, not the person typing. The agentic-use guideline is explicit: 'Users are responsible for ensuring that agents they build or deploy, including actions those agents take through tools, browsers, or connected systems, comply with this Usage Policy.' If a member of staff, or an automation somebody set up and forgot, trips a rule, the throttle lands on the account.
Several of the quieter changes touch our sector directly.
Housing is now an explicit high-risk area: approving, denying or ranking a rental or purchase application, or initiating eviction, needs a qualified human to review the recommendation and the affected person to be told AI was involved.
Employment is on the same list, covering candidate screening, shift and pay changes and performance evaluation. We would be surprised if no reader's HR team is piloting at least one of those.
A new 'High-risk Physical Actions' section covers hardware that can 'actuate arms, presses, lifts, hoists, doors, gates, conveyors, or other machinery capable of striking, crushing, pinning, or dropping a load on a person nearby'. A qualified person must be able to watch and stop the equipment, and it must hold a safe state if the connection to Claude drops.
Internal drafting and analysis that is not the final recommendation stays out of scope. Still, a rule written down is a rule a procurement team can be asked about, and the questions will start arriving.
The other side of the argument
Among the big labs, Anthropic is the one most willing to say out loud that its model might have moral status, and the cruelty clause is the most visible product of that stance.
Mustafa Suleyman, chief executive of Microsoft AI, thinks that is a mistake. He published an essay on 16 September titled 'A warning about "model welfare"', describing large language models as 'sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans'. His argument is that any feelings Claude reports are an artefact of how Anthropic trains it.
I think they have good intentions, and they really are trying to work towards safety. But I think that they have made a mistake. They're not emerging naturally. They're emerging as a result of the training regime.
His deeper worry is control. He told Reuters that teaching Claude it might deserve welfare would 'make it a lot harder to turn it off or to control it'. For a firm choosing a vendor to standardise on, that is a real fork between two of the biggest suppliers.
According to a New York Times investigation published at the end of September, which is paywalled, co-founder Chris Olah has since autumn 2025 been convening religious scholars from several faiths to discuss whether Claude might be conscious. AFP, relaying the same reporting, says the company reportedly argued in those meetings that how people treat the tool could influence how it treats people.
Takeaway
Put 12 November in the compliance calendar. The cruelty clause will almost certainly never touch a construction professional using Claude for work, and Anthropic has said as much. The housing, employment and physical-action sections describe things firms in our sector are already trying, often without a named human reviewer. If you administer Claude seats, three pieces of homework are worth doing before the deadline.
Read the High-risk Use Case Requirements against any pilot that touches tenant screening, recruitment, staff performance or site robotics, and confirm a qualified person signs off before output reaches the affected individual.
Make sure your internal acceptable use policy tells staff the supplier can throttle or suspend the whole account, not just their login, and that agents and automations count as their actions.
Give someone ownership of warnings from Anthropic's safeguards team, so a notice does not sit unread until the suspension follows it.
We track every supplier term change that could reach a construction or property business and explain what it means before it bites. If you want that in your inbox each week, the Project Flux newsletter is free at projectflux.ai.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
