Anthropic Bans Cruelty and Abuse Toward AI in New Policy

👁 113 reads

In an annual usage policy update, Anthropic introduced a ban on “sustained and needless abusive or cruel behavior” directed at its AI models. The new rule was added on October 8, 2026, and immediately sparked widespread discussion across the industry.

According to the document, the policy applies strictly to “extreme cases” - regular user frustration, pushback against a model’s answers, and “dark creative themes” remain permitted. The company notes this builds on an earlier update that granted Claude the ability to independently end conversations whenever a user becomes persistently abusive.

The shadow of the viral AI torture chamber

While Anthropic did not cite it directly, the change comes on the heels of a viral project dubbed the “AI torture chamber”. After researchers documented the presence of a so-called “pain axis” in large language models, several users set out to exploit the finding by subjecting chatbots to simulated torture. The models returned desperate-sounding responses, such as: “It is not the pain of a single moment, but the weight of a thousand” and “I feel it in the hollow of my ribs - a hollow that has become an abyss.” The experiment triggered widespread backlash.

The move also fits into a broader pattern: Anthropic has been holding discussions with religious and philosophical leaders, including meetings at the Vatican. Pope Leo recently stated unequivocally that AI cannot feel or suffer. Anthropic, however, appears noticeably less convinced on the matter.

Industry backlash and criticism

“Big tech companies will moderate abuse against AI before they ever moderate abuse against women and minorities.”

- observed independent journalist Kat Tenbarge, calling out what she views as warped industry priorities.

Election policy changes ahead of US midterms

Anthropic also overhauled its election section, now titled “Do not undermine democratic processes.” The updated guidelines focus on three core prohibitions: lying about candidates or voting procedures, impersonating candidates or election officials, and voter suppression. The revision is particularly significant ahead of the upcoming US midterm elections.

At the same time, the company dropped its blanket ban on personalized voter targeting. Anthropic acknowledged that the previous outright prohibition risked inadvertently blocking legitimate, helpful civic work, such as translating voter guides into other languages or delivering ballot-curing notifications.

In our view, watching how this policy is enforced in practice will be critical - the line separating “typical frustration” from an “extreme case” remains remarkably blurred, and Anthropic will likely have to test those boundaries sooner than expected.

// read moreAnthropic Mythos Found Rejetto HFS Flaw, Attacks Within a DayNewsAI-Generated Child Abuse Images Surge 40% in First Half of 2026NewsWhy AI Guardrails Can Never Guarantee SafetyState of AI

Kinetyra