Anthropic changes usage policy to ban model abuse and election interference
News Source
β’Thu, 08 Oct 2026 18:16:24 +0000
π° What Happened
Anthropic has updated its usage policy with new rules. It bans using its Claude models for election interference, weapons software and surveillance. It also forbids users from abusing the AI for long periods.
The company says the abuse rule only applies in extreme cases. Normal frustration, dark creative writing and testing are still allowed. A separate section bars deception, such as running fake accounts or fake news outlets.
π The Backstory
Anthropic added training in August so Claude ends harmful conversations. The new policy goes further by banning users from starting them. The company has also talked with religious scholars about whether Claude could be conscious.
The update comes as AI firms face pressure to set clear rules. How their models are used in elections and warfare has become a major concern. Many companies are now writing stricter policies.
π― Why It Matters
These rules shape what AI chatbots will and will not do. They affect how safe the tools are during elections and in everyday use.
Anthropic has updated its usage policy with new rules. It bans using its Claude models for election interference, weapons software and surveillance. It also forbids users from abusing the AI for long periods.
The company says the abuse rule only applies in extreme cases. Normal frustration, dark creative writing and testing are still allowed. A separate section bars deception, such as running fake accounts or fake news outlets.
Anthropic added training in August so Claude ends harmful conversations. The new policy goes further by banning users from starting them. The company has also talked with religious scholars about whether Claude could be conscious.
The update comes as AI firms face pressure to set clear rules. How their models are used in elections and warfare has become a major concern. Many companies are now writing stricter policies.
These rules shape what AI chatbots will and will not do. They affect how safe the tools are during elections and in everyday use.